JP4310847B2 - Image information conversion apparatus and conversion method - Google Patents

Description

【００２０】
ここで、ｎはクラスタップ中の画素数であり、図２に示すようなクラスタップのタップ構造においてはｎ＝７である。また、ｐの値としては、例えばｐ＝２とすることができる。
【００２１】
また、領域切り出し回路６が切り出す予測タップのタップ構造の一般的な例を図３に示す。ここで、水平方向は時間の推移を示す。すなわち、＃（ｋ−２），＃（ｋ−１），‥‥，＃（ｋ＋２）は、それぞれ、時点ｋ−２，ｋ−１，‥‥，ｋ＋２におけるフィールドを示している。また、垂直方向は各フィールドにおける垂直方向を示す。この例では、ｋ₁ 〜ｋ₇ の７個の画素が予測タップとされる。
【００２２】
ｈｄ'＝ｗ₁×ｘ₁＋ｗ₂×ｘ₂＋‥‥＋ｗ_n×ｘ_n （３）
ここで、ｘ₁，‥‥，ｘ_nが各予測タップの画素データであり、ｗ₁，‥‥，ｗ_nが各予測係数セットである。
【００２３】
ｈｄ’＝ｗ₁ ×ｘ₁ ＋ｗ₂ ×ｘ₂ ＋‥‥＋ｗ_n ×ｘ_n （３）
ここで、ｘ₁ ，‥‥，ｘ_n が各予測タップの画素データであり、ｗ₁ ，‥‥，ｗ_n が各予測係数である。
【００２４】
上述したようにしてクラス分類適応処理によってフィールド倍速画像を生成する場合には、ＳＤ画像データを単に補間処理する場合に比べ、より精度良くフィールド倍速画像を生成することができる。なお、予測係数は、ＳＤ画像データ内の画素位置と、生成されるべきフィールド倍速画像内の画素位置との関係に応じて、後述するような処理によって決定される。
【００２５】
次に、ＳＤ画像としての５０Ｈｚ画像内の画素位置とフィールド倍速画像としての１００Ｈｚ画像内の画素位置との位置関係に応じた変換性能について説明する。まず、本願出願人によるフィールド倍速処理に係る先の提案（９８００２３７５０６参照）における、創造すべきフィールドの画素位置と予測タップの画素位置との関係の例を図４に示す。この場合、創造される１００Ｈｚ画素ｍ₁ ，ｍ₂ ，ｍ₃ ，ｍ₄ の内、ｍ₁ は時間的空間的に５０Ｈｚ画素と一致した位置にあるため、ｍ₁ の創造に係る変換性能は極めて良好となる。
【００２６】
これに対し、ｍ₂ およびｍ₄ は、５０Ｈｚ画素との時間的な隔たりが大きいので、ｍ₂ およびｍ₄ の創造に係る変換性能は特に動画像においてｍ₁ の創造に係る変換性能に比べて劣るという問題があった。一般に、画像全体についての画質の良否は、当該画像全体の中で最も劣化した部分に強く影響される傾向がある。このため、画像全体についての画質の向上を図る上で、画素の位置関係によって変換性能が大きく異なることは好ましくない。
【００２７】
また、１００Ｈｚ画素ｍ₂ およびｍ₄ は、時間的な対称性を利用すると互いに同一の予測係数を使用することによって創造することができる。これに対して、１００Ｈｚ画素ｍ₁ およびｍ₃ を創造するためにはそれぞれ別個の予測係数を使用する必要がある。すなわち、図４に示した例においては３種類の予測係数が必要となる。このため、ハードウエアの規模が大きくなるので、特にハードウエアの規模が制約される場合において問題となることがある。
【００２８】
そこで、この発明は、ＳＤ画像内の画素とフィールド倍速画像内の画素との画素位置関係としてより適切なものを選択できるようにしたものである。以下、図４に示した例と共に、図５、図６、図７にそれぞれ示す、この発明の一実施形態において用いることが可能な画素位置関係の一例、他の例、さらに他の例について順次説明する。
【００２９】
図４に示した画素位置関係の例を用いてフィールド倍速変換を行う場合には、ｍ₁ の予測精度が最も高く、ｍ₃ の予測精度がそれに次いで高く、ｍ₂ およびｍ₄ の予測精度が最も低い。そして、変換によって得られる画像全体の画質は、ｍ₂ およびｍ₄ の予測精度に大きく影響される。また、上述したように、全部で３種類の予測係数が必要とされる。また、この画素位置関係においては、ラインフリッカが強調されることは無い。
【００３０】
図６に示す画素位置関係の他の例を用いてフィールド倍速変換を行う場合には、ｍ₁，ｍ₂，ｍ₃，ｍ₄の４種類の画素の各々について予測精度が異なる。但し、図４に示した例と比較するとばらつきは小さく、ばらつきの程度は図５に示した一例と略同等となる。また、ｍ₁およびｍ₃を創造するに際しては、時間方向の対称性を利用することによって予測係数の共通化を行うことができる。さらに、ｍ₂およびｍ₄を創造するに際しては、空間（垂直）方向の対称性を利用することによって予測係数の共通化を行うことができる。従って、創造すべき４種類の画素について全部で２種類の予測係数が必要となる。また、この場合にはラインフリッカが強調されることは無い。
【００３１】
図６に示す画素位置関係の他の例を用いてフィールド倍速変換を行う場合には、ｍ₁ ，ｍ₂ ，ｍ₃ ，ｍ₄ の４種類の画素の各々について予測精度が異なる。但し、図４に示した例と比較するとばらつきは小さく、ばらつきの程度は図５に示した一例と略同等となる。また、ｍ₁ およびｍ₃ を創造するに際しては、時間方向の対称性を利用することによって予測係数の共通化を行うことができる。さらに、ｍ₂ およびｍ₄ を創造するに際しては、空間（垂直）方向の対称性を利用することによって予測数の共通化を行うことができる。従って、創造すべき４種類の画素について全部で２種類の予測係数が必要となる。また、この場合にはラインフリッカが強調されることは無い。
【００３２】
図７に示す画素位置関係のさらに他の例を用いてフィールド倍速変換を行う場合には、ｍ₁ ，ｍ₂ ，ｍ₃ ，ｍ₄ の４種類の画素の各々について予測精度が同一となり、予測精度の画素位置によるばらつきは無い。また、時間方向の対称性／空間（垂直）方向の対称性を利用することによってｍ₁ 〜ｍ₄ の全ての予測係数の共通化を行うことができる。従って、で創造すべき４種類の画素の全てを、１種類の予測係数の下で創造することができる。また、この場合にはラインフリッカが強調されるおそれがある。
【００３３】
上述したような画素位置関係の各例についての特徴を考慮して、予測係数を保持するためのメモリ容量等のハードウエア的な制約、ラインフリッカが許容されるか否か等の用途等に係る条件等に応じて最も好適な画素位置関係を選ぶことにより、総合的な変換性能を向上させることができる。
【００３４】
次に、この発明の一実施形態における学習、すなわち予測係数を決定する処理について図８を参照して説明する。入力端子２０を介して、画像情報変換の結果として生成されるべき画像と同一の信号形式を有する画像（教師画像と称される）内の画素を全て含む画像が供給される。この発明の一実施形態は、５０Ｈｚ画像を１００Ｈｚ画像に変換するものなので、１００Ｈｚプログレッシブ画像が入力端子２０を介して垂直位相シフトフィルタ２１および遅延回路２８に供給される。垂直位相シフトフィルタ２１は、図４〜図７を参照して上述したような画素位置関係の何れを前提として学習を行うかに応じた処理を行う（これについては後述する）。
【００３５】
垂直位相シフトフィルタ２１の出力が時間間引きフィルタ２２に供給される。時間間引きフィルタ２２は、垂直位相シフトフィルタ２１の出力に対して、時間方向に平滑化する処理を施すことにより、５０Ｈｚインターレース画素位置に画素を生成する。このようにして、ＳＤ画像と同一の信号形式（すなわち５０Ｈｚインターレース画像）を生成することができる。かかる画像は生徒画像と称される。生徒画像が領域切り出し回路２３、２６に供給される。
【００３６】
領域切り出し回路２３は、図１中の領域切り出し回路２と同一の処理を行う。すなわち、領域切り出し回路２３は、供給される生徒画像から、領域切り出し回路２によって切り出されるものと同一の画素領域をクラスタップとして切り出し、切り出したクラスタップのデータをＡＤＲＣ回路２４に供給する。ＡＤＲＣ回路２４は、供給されるデータに図１中のＡＤＲＣ回路３と同一の処理を施す。すなわち、上述の式（１）に示したような演算を行うことにより、例えば８ビットを単位とするデータから２ビットを単位とする再量子化コードを形成する。形成された再量子化コードがパターン圧縮データとしてクラスコード発生回路２５に供給する。
【００３７】
クラスコード発生回路２５は、図１中のクラスコード発生回路３と同一の処理を行う。すなわち、クラスコード発生回路２５は、ＡＤＲＣ回路２４の出力に基づいて上述の式（２）に示したような演算を行うことにより、クラスコードを発生する。クラスコード発生回路２５の出力は、正規方程式加算回路２９に供給される。一方、領域切り出し回路２６は、図１中の領域切り出し回路６と同一の処理を行う。すなわち、領域切り出し回路２６は、供給される画像信号から、領域切り出し回路６によって切り出されるものと同一の画素領域を予測タップとして切り出し、切り出した予測タップのデータを遅延回路２７に供給する。遅延回路２７は、供給されるデータを一旦保持し、その後、正規方程式加算回路２９に供給する。
【００３８】
また、遅延回路２８は、供給される１００Ｈｚプログレッシブ画像を一旦保持し、その後、正規方程式加算回路２９に供給する。遅延回路２８および２９により、算処理を行ために好適なタイミングで、正規方程式加算回路２９が領域切り出し回路２６の出力および１００Ｈｚプログレッシブ画像を供給されることが担保される。
【００３９】
正規方程式加算回路２９は、遅延回路２７から供給されるクラスタップのデータ、遅延回路２５から供給されるクラスコード、および遅延回路２８から供給される１００Ｈｚプログレッシブ画像に基づいて、後述するような正規方程式のデータを算出する。ここで、正規方程式加算回路２９は、１００Ｈｚプログレッシブ画像から教師画像を構成するデータを抽出し、抽出したデータに基づく演算処理を行う。算出されるデータは、予測係数決定回路３０に供給される。予測係数決定回路３０は、供給されるデータに基づいて正規方程式を解くための演算を行う。かかる演算の結果として算出された予測係数はメモリ３１に供給される。図１を参照して上述した構成による処理を行うに先立ち、メモリ３１の記憶内容が係数メモリ５にロードされる。
【００４０】
予測係数を算出するための演算処理について説明する。以下の説明では、予測タップとして切り出される画素数を一般化してｎとする。すなわち、ｘ₁ ，ｘ₂ ，‥‥，ｘ_n を予測タップとする場合、フィールド倍速画像中の画素レベルｙが次の式（４）に従って計算されるとする。
【００４１】
ｙ＝ｗ₁ ×ｘ₁ ＋ｗ₂ ×ｘ₂ ＋‥‥＋ｗ_n ×ｘ_n （４）
学習前は、予測係数ｗ₁ ，‥‥，ｗ_n が未定係数である。学習は、クラス毎に複数の画像データについて行う。画像データの種類数をｍと表記する場合、式（４）から、以下の式（５）が設定される。
【００４２】
ｙ_k ＝ｗ₁ ×ｘ_k1＋ｗ₂ ×ｘ_k2＋‥‥＋ｗ_n ×ｘ_kn （５）
（ｋ＝１，２，‥‥，ｍ）
ｍ＞ｎの場合には、ｗ₁ ，‥‥，ｗ_n は一意に決まらないので、誤差ベクトルｅの要素ｅ_k を以下の式（６）で定義して、式（７）によって定義される誤差ベクトルｅを最小とするように予測係数を定めるようにする。すなわち、いわゆる最小２乗法によって予測係数を一意に定める。
【００４３】
ｅ_k ＝ｙ_k −｛ｗ₁ ×ｘ_k1＋ｗ₂ ×ｘ_k2＋‥‥＋ｗ_n ×ｘ_kn｝ （６）
（ｋ＝１，２，‥‥ｍ）
【００４４】
【数２】

【００４５】
式（７）のｅ² を最小とする予測係数を求めるための実際的な計算方法としては、ｅ² を予測係数ｗ_i (i=1,2‥‥）で偏微分し（式（８））、ｉの各値について偏微分値が０となるように各予測係数ｗ_i を決定すれば良い。
【００４６】
【数３】

【００４７】
式（８）から各予測係数ｗ_i を決定する具体的な手順について説明する。式（９）、（１０）のようにＸ_ji，Ｙ_i を定義すると、式（８）は、式（１１）の行列式の形に書くことができる。
【００４８】
【数４】

【００４９】
【数５】

【００５０】
【数６】

【００５１】
式（１１）が一般に正規方程式と呼ばれるものである。正規方程式加算回路２９は、供給されるデータに基づいて式（９）、（１０）に示すような演算を行うことにより、Ｘ_ji，Ｙ_i （ｉ＝１，２，‥‥，ｎ）をそれぞれ計算する。予測係数決定回路３０は、掃き出し法等の一般的な行列解法に従って正規方程式（１１）を解くことにより、予測係数ｗ_i （ｉ＝１，２，‥‥，ｎ）を算出する。以上のような処理の結果として、統計的に最も真値に近い推定値を各クラス毎のフィールド倍速データｙとして推定するための予測係数が算出される。
【００５２】
この発明は、上述したような学習を、複数の画素位置関係の各々について選択的に行うようにしたものである。この発明の一実施形態では、図４〜図７を参照して上述した４種類の画素位置関係の何れに対しても、対応する予測係数を計算できるように構成されている。まず、図５および図７を参照して上述したような画素位置関係においては、入力する５０Ｈｚ画像内の画素と、創造すべき１００Ｈｚ画像内の画素との垂直方向の画素位置がずれている。かかる位置関係に対応した学習を行う際には、垂直間位相シフトフィルタ２１は、供給される１００Ｈｚプログレッシブ画像に対して垂直方向の位相シフトを行う。
【００５３】
一方、図４および図６を参照して上述したような画素位置関係においては、入力する５０Ｈｚ画像内の画素と、創造すべき１００Ｈｚ画像内の画素との垂直方向の画素位置が一致している。かかる位置関係に対応した学習を行う際には、垂直間引きフィルタ２１は、位相シフトは行わず、供給される１００Ｈｚプログレッシブ画像をそのまま出力する。
【００５４】
上述したようにして、画像情報変換の対象および結果である、５０Ｈｚ画像および１００Ｈｚ画像の間の複数種類の画素位置関係の内で、画像情報変換の結果として生成されるべき１００Ｈｚ画像のとしてより高品質の画像が生成されるような学習を行うことが可能となる。
【００５５】
また、予測係数を保持するためのメモリ容量等のハードウエア的な制約、ラインフリッカが許容されるか否か等の用途等に係る条件等に関連して最も好適な画素位置関係に対応する学習を選択的に行うことができる。これにより、より好適な画素位置関係を前提とするクラス分類適応処理を適用してフィールド倍速変換等の画像情報変換を行うことが可能となるので、総合的な変換性能を向上させることができる。
【００５６】
上述したこの発明の一実施形態は、５０Ｈｚ画像を１００Ｈｚ画像に変換するフィールド倍速変換を行うものである。これに対して、学習時に、例えば垂直方向の位相シフト等を必要に応じて行うことによって教師画像、或いは生徒画像を生成することにより、インターレース画像をプログレッシブ画像に変換する、或いはＳＤ画像をＨＤ(High Definition) 画像に変換する等の画像情報変換を行う場合等においても、この発明を適用することができる。すなわち、この発明は、ディジタル画像を、より画素数の多いディジタル画像に変換する画像情報変換全般において有効である。
【００５７】
また、この発明の一実施形態は、１００Ｈｚのプログレッシブ画像と５０Ｈｚのインターレース画像との間で学習を行うものであるが、学習に係る画像の組み合わせはこれに限定されるものではない。例えば、５０Ｈｚのプログレッシブ画像と２５Ｈｚのインターレース画像との間で学習を行うようにしても良いし、６０Ｈｚのプログレッシブ画像と３０Ｈｚのインターレース画像との間で学習を行うようにしても良い。
【００５８】
また、上述したこの発明の一実施形態では、５０Ｈｚ画像等の入力画像の空間波形を少ないビット数でパターン化するためにＡＤＲＣを行うようにしている。これに対して、時空間内における画像パターンを少ないクラスで表現することに資するような情報圧縮方法であれば、ＡＤＲＣ以外の方法を用いるようにしても良い。例えば、ＤＰＣＭ(Differential Pulse Code Modulation)や、ＶＱ（Vector Quantization ）等の方法を用いるようにしても良い。
【００５９】
この発明は、上述したこの発明の一実施形態に限定されるものでは無く、この発明の主旨を逸脱しない範囲内で様々な変形や応用が可能である。
【００６０】
【発明の効果】
この発明によれば、生徒画像と教師画像との複数種類の組合わせの内から何れかの組合わせを選択し、選択した組合わせの下で学習を行うことによって決定される予測係数を使用して予測演算を行うことができる。
【００６１】
このため、画像情報変換の結果としてより高品質の画像が生成されるような学習を行うことが可能となる。
【００６２】
また、予測係数を保持するためのメモリ容量等のハードウエア的な制約、ラインフリッカが許容されるか否か等の用途等に係る条件等に関連して最も好適な画素位置関係に対応する学習を行うことが可能となる。
【００６３】
また、生徒画像と教師画像との組合わせによって時間方向或いは空間方向の対称性が異なることに起因して、予測係数の種類数も異なるので、予測係数として記憶されるべきデータ量は、選択される生徒画像と教師画像との組合わせによって異なる。このため、この発明を適用することにより、装置の仕様等において許容される予測係数用のメモリの容量等の条件により適合する画像情報変換を行うことができる。
【図面の簡単な説明】
【図１】この発明の一実施形態における画像情報変換に係る構成の一例を示すブロック図である。
【図２】クラスタップのタップ構造の一例を示す略線図である。
【図３】予測タップのタップ構造の一例を示す略線図である。
【図４】本願出願人の先の提案に係る画像情報変換における画素位置関係について説明するための略線図である。
【図５】この発明の一実施形態において用いることができる画素位置関係の一例について説明するための略線図である。
【図６】この発明の一実施形態において用いることができる画素位置関係の他の例について説明するための略線図である。
【図７】この発明の一実施形態において用いることができる画素位置関係のさらに他の一例について説明するための略線図である。
【図８】この発明の一実施形態における学習について説明するためのブロック図である。
【符号の説明】
２１・・・垂直位相シフトフィルタ[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image information conversion apparatus and conversion method for converting an input digital image signal into a digital image signal having a larger number of pixels.
[0002]
[Prior art]
Conventionally, as a scanning method for a television receiver, an interlace scanning method in which scanning is performed by skipping every other horizontal scanning line (line) has been most widely adopted. In the interlace scanning method, one frame is formed by two fields of an odd field obtained by scanning odd lines and an even field obtained by scanning even lines.
[0003]
The interlace scanning method is adopted as a television standard method in various countries around the world. For example, in European television broadcasting, an interlace scanning method with a field frequency of 50 Hz is used.
[0004]
[Problems to be solved by the invention]
In a television receiver using an interlace scanning method with a field frequency of 50 Hz, there is a problem that flicker called surface flicker occurs over a wide range of the display screen. The surface flicker occurs due to the fact that each line is displayed at every predetermined time interval because it takes a predetermined time to switch from the odd field to the even field in the interlace scanning method.
[0005]
As a method for coping with such a problem, a field double speed process for generating an image signal of, for example, 100 Hz by generating an interpolation field by a median filter based on an input (for example, 50 Hz SD (Standard Definition)) image is known. ing. This method uses an intermediate pixel value as a pixel value of an interpolated pixel in, for example, three pixels close to the interpolated pixel in a field including the interpolated pixel and in a field preceding the field including the interpolated pixel. It is what.
[0006]
However, in such a method, when the SD image to be subjected to the interpolation process is a still image, unnatural distortion occurs in an image region having a high vertical resolution. In addition, when the SD image to be interpolated is a moving image, the pixel value of the pixel one field before may be selected as the pixel value of the interpolation pixel. When such a selection is made, There was a problem that unnatural movement occurred.
[0007]
In order to solve such a problem, the applicant of the present application has proposed a method of converting an input image into a field double speed image by performing a classification adaptation process. However, in such a method, there is a possibility that the conversion performance partially varies depending on the positional relationship between the pixel in the image to be generated and the pixel in the SD image, and as a result, improvement in the image quality of the entire field double speed image may be hindered. there were.
[0008]
Accordingly, an object of the present invention is to perform image information conversion such as a field double-speed image by performing processing assuming a plurality of types of pixel positional relationships between pixels in an image to be generated and pixels in an SD image. It is an object of the present invention to provide an image information conversion apparatus and conversion method capable of improving the image quality of an image generated as a result of the above.
[0009]
[Means for Solving the Problems]
The invention of claim 1 Interlaced scanning By generating a new field between the fields of the input digital image signal, the number of fields greater than the number of fields of the input digital image signal Interlaced scanning In an image information conversion apparatus that forms an output digital image signal, a first image cutout unit that cuts out a plurality of image data around a target pixel to be processed from the input digital image signal, and a plurality of pieces cut out by the first image cutout unit Detecting a level distribution of each of the image data, determining a class to which the target pixel belongs from the detected level distribution, outputting class detection information representing the determined class, and a target pixel from the input digital image signal A second image cutout unit that cuts out a plurality of image data around the image, and a prediction coefficient that is determined in advance corresponding to each of the classes and stores an output digital image signal, and stores the prediction coefficient from among the stored prediction coefficients Coefficient storage means for outputting a prediction coefficient corresponding to the class detection information from the class detection means; An arithmetic processing unit that performs a product-sum operation on a plurality of pieces of image data cut out by the image cutout unit and a prediction coefficient supplied from the coefficient storage unit, and generates a predicted value of the image data of the output digital image signal; The prediction coefficient is obtained by learning in advance for each class and stored in the coefficient storage means. The learning is performed by calculating the pixel position of the input digital image signal. In contrast, an output digital image signal having a first positional relationship that includes a field that matches the field of the input digital image signal in the time direction and that has a pixel position in the vertical direction that matches the input digital image signal; And an output digital image signal having a second positional relationship in which the pixel position is symmetrical in the vertical direction with the pixel position of the input digital image signal as the center, and a field of the input digital image signal The field is symmetrical in the time direction and the output digital image signal in the third positional relationship in which the pixel position in the vertical direction coincides with the input digital image signal and the field of the input digital image signal is in the time direction. And the pixel position of the input digital image signal Pixel position as the center of the fourth positional relation of symmetry in the vertical direction Output digital image signal Of the output digital image signal of the first, second, third and fourth positional relationships One from the inside Position-related output digital image signal Select The input digital image signal is the pixel of the teacher image signal in the same signal format as the selected output digital image signal. To the pixel position of conversion A digital image signal in which the number of fields per unit time of the output digital image signal is converted so as to be the same as the input digital image signal, Generated From digital image signal chosen Output digital image signal Pixel value of Generate a Do Sometimes generated Be done Pixel value and teacher In order to minimize the error from the true value of the pixel at the corresponding position of the image signal, The prediction factor This is an image information conversion apparatus which is a process to be obtained.
[0010]
Claim 4 The invention of Interlaced scanning By generating a new field between the fields of the input digital image signal, the number of fields greater than the number of fields of the input digital image signal Interlaced scanning In an image information conversion method for forming an output digital image signal, a first image cutout step of cutting out a plurality of image data around a target pixel to be processed from the input digital image signal, and a plurality of cutouts by the first image cutout step Detecting a level distribution of each of the image data, determining a class to which the target pixel belongs from the detected level distribution, outputting class detection information expressing the determined class, and a target pixel from the input digital image signal A second image cut-out step for cutting out a plurality of image data around the image, and a prediction coefficient determined in advance corresponding to each of the classes and storing an output digital image signal, from among the stored prediction coefficients , Output the prediction coefficient corresponding to the class detection information from the class detection step A product-sum operation is performed on the plurality of pieces of image data cut out by the number storage step, the second image cut-out step, and the prediction coefficient supplied from the coefficient storage step, and a predicted value of the image data of the output digital image signal is generated. A prediction coefficient is obtained by learning in advance for each class and stored, and learning is performed by calculating the pixel position of the input digital image signal. In contrast, an output digital image signal having a first positional relationship that includes a field that matches the field of the input digital image signal in the time direction and that has a pixel position in the vertical direction that matches the input digital image signal; And an output digital image signal having a second positional relationship in which the pixel position is symmetrical in the vertical direction with the pixel position of the input digital image signal as the center, and a field of the input digital image signal The field is symmetrical in the time direction and the output digital image signal in the third positional relationship in which the pixel position in the vertical direction coincides with the input digital image signal and the field of the input digital image signal is in the time direction. And the pixel position of the input digital image signal Pixel position as the center of the fourth positional relation of symmetry in the vertical direction Output digital image signal Of the output digital image signal of the first, second, third and fourth positional relationships One from the inside Position-related output digital image signal Select The input digital image signal is the pixel of the teacher image signal in the same signal format as the selected output digital image signal. To the pixel position of conversion A digital image signal in which the number of fields per unit time of the output digital image signal is converted so as to be the same as the input digital image signal, Generated From digital image signal chosen Output digital image signal Pixel value of Generate a Do Sometimes generated Be done Pixel value and teacher In order to minimize the error from the true value of the pixel at the corresponding position of the image signal, The prediction factor This is an image information conversion method that is a process to be obtained.
[0011]
According to the invention as described above, a prediction coefficient can be generated using a combination more suitable for desired image information conversion among a plurality of types of combinations of a student image signal and a teacher image signal. .
[0012]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows an example of a configuration relating to image information conversion in one embodiment of the present invention. An image signal obtained by digitizing a 50 Hz interlaced image in the PAL system is supplied to the area clipping circuits 2 and 6 through the input terminal 1 as SD image data. The area extraction circuit 2 extracts a predetermined pixel area from the supplied SD image data as a class tap, and supplies the extracted class tap data to an ADRC (Adaptive Dynamic Range Coding) circuit 3. The ADRC circuit 3 converts the supplied data into A The DRC process is performed, and the code value generated thereby is supplied to the class code generation circuit 4.
[0013]
The class code generation circuit 4 generates a class code indicating a class corresponding to a pattern in space-time based on the supplied code value. The generated class code is supplied to the coefficient memory 5. The coefficient memory 5 is determined in advance as will be described later. Forecast Coefficient of measurement set Is stored corresponding to the class code.
[0014]
On the other hand, the region cutout circuit 6 cuts out a predetermined pixel region from the supplied 50 Hz interlaced image as a prediction tap, and supplies the cutout tap data to the delay circuit 7. The delay circuit 7 once holds the supplied data, and supplies the held data to the estimation arithmetic circuit 8 at an appropriate timing. Thereby, the timing at which the prediction tap data is supplied to the estimation calculation circuit 8 and the timing at which the prediction coefficient is supplied from the coefficient memory 5 to the estimation calculation circuit 8 are matched. The estimation operation circuit 8 generates a 100 Hz image as a field double speed image by performing an operation based on the output of the delay circuit 7 and the output of the coefficient memory 5. The generated 100 Hz image is displayed on a display device such as a television receiver (not shown) via the output terminal 9. etc To be supplied.
[0015]
Hereinafter, processing by each component will be described in detail. A typical example of the tap structure of the class tap cut out by the area cutout circuit 2 is shown in FIG. Here, the horizontal direction shows the transition of time. That is, # (k−2), # (k−1),..., # (K + 2) indicate fields at time points k−2, k−1,. The vertical direction indicates the vertical direction in each field. In a 50 Hz interlaced image, the scanning line phase is shifted between adjacent fields. In this example, k ₁ ~ K ₇ These seven pixels are class taps.
[0016]
The ADRC circuit 3 converts, for example, 8-bit class tap pixel data per pixel into a 2-bit requantization code value. The level quantization pattern in the class tap data is expressed with a smaller amount of information by the requantization code value generated in this way. That is, the output of the ADRC circuit 3 is used as pattern compression data. ADRC is an adaptive requantization method originally developed for high-efficiency coding for VTR (Video Tape Recorder), but it can efficiently express local patterns of signal level in short tone. Has characteristics. Therefore, the pattern in the space-time of the image signal to generate the classification code In Can be used. The ADRC circuit 3 equally divides the maximum value MAX and the minimum value MIN in the area cut out as a class tap by the specified number of bits according to the following equation (1). , The data cut out as a class tap Requantize.
[0017]
DR = MAX-MIN + 1
Q = [(L−MIN + 0.5) × 2 ⁿ / DR] (1)
Here, DR is a dynamic range within the region. Also, n is the number of bits allocated, and for example, n = 2 can be set. L is the data level of the pixels in the area, and Q is the requantization code. However, the brackets ([...]) Indicate truncation processing.
[0018]
K as shown in FIG. ₁ ~ K ₇ When the tap structure of class taps consisting of 7 pixels is used, the 7 requantization codes q corresponding to each pixel are obtained by the processing as described above. ₁ ~ Q ₇ Is the output of the ADRC circuit 3. The class code class generated by the class code generation circuit 4 based on such output is as shown in the following equation (2).
[0019]
[Expression 1]

[0020]
Here, n is the number of pixels in the class tap, and n = 7 in the tap structure of the class tap as shown in FIG. Moreover, as a value of p, it can be set as p = 2, for example.
[0021]
A general example of the tap structure of the prediction tap cut out by the area cutout circuit 6 is shown in FIG. Here, the horizontal direction shows the transition of time. That is, # (k−2), # (k−1),..., # (K + 2) indicate fields at time points k−2, k−1,. The vertical direction indicates the vertical direction in each field. In this example, k ₁ ~ K ₇ These seven pixels are taken as prediction taps.
[0022]
hd '= w ₁ X ₁ + W ₂ X ₂ + ... + w _n X _n (3)
Where x ₁ , ..., x _n Is the pixel data of each prediction tap, and w ₁ , ..., w _n Is each prediction coefficient set It is.
[0023]
hd '= w ₁ X ₁ + W ₂ X ₂ + ... + w _n X _n (3)
Where x ₁ , ..., x _n Is the pixel data of each prediction tap, and w ₁ , ..., w _n Is each prediction coefficient.
[0024]
As described above, when the field double speed image is generated by the class classification adaptive process, the field double speed image can be generated with higher accuracy than when the SD image data is simply interpolated. Note that the prediction coefficient is determined by processing as will be described later in accordance with the relationship between the pixel position in the SD image data and the pixel position in the field double speed image to be generated.
[0025]
Next, the conversion performance according to the positional relationship between the pixel position in the 50 Hz image as the SD image and the pixel position in the 100 Hz image as the field double speed image will be described. First, FIG. 4 shows an example of the relationship between the pixel position of the field to be created and the pixel position of the prediction tap in the previous proposal (see 9800237506) related to the field double speed processing by the applicant of the present application. In this case, the created 100 Hz pixel m ₁ , M ₂ , M _Three , M _Four Of which m ₁ Is in a position that coincides with the 50 Hz pixel in time and space, m ₁ The conversion performance related to the creation of is very good.
[0026]
In contrast, m ₂ And m _Four Has a large temporal separation from the 50 Hz pixel. ₂ And m _Four The conversion performance related to the creation of ₁ There was a problem that it was inferior to the conversion performance related to the creation. In general, the quality of an image as a whole tends to be strongly influenced by the most deteriorated portion of the entire image. For this reason, in order to improve the image quality of the entire image, it is not preferable that the conversion performance varies greatly depending on the positional relationship of the pixels.
[0027]
Also, 100Hz pixel m ₂ And m _Four Can be created by using the same prediction coefficients when using temporal symmetry. In contrast, 100 Hz pixel m ₁ And m _Three In order to create each, it is necessary to use a separate prediction coefficient. That is, in the example shown in FIG. 4, three types of prediction coefficients are required. For this reason, since the scale of hardware becomes large, it may become a problem especially when the scale of hardware is restricted.
[0028]
Therefore, the present invention enables selection of a more appropriate pixel positional relationship between the pixels in the SD image and the pixels in the field double speed image. Hereinafter, in addition to the example shown in FIG. 4, one example of the pixel positional relationship that can be used in one embodiment of the present invention, another example, and another example shown in FIG. 5, FIG. 6, and FIG. explain.
[0029]
When performing field double speed conversion using the example of the pixel position relationship shown in FIG. ₁ Has the highest prediction accuracy of m _Three Has the next highest prediction accuracy, m ₂ And m _Four Has the lowest prediction accuracy. The image quality of the entire image obtained by the conversion is m ₂ And m _Four Is greatly affected by the prediction accuracy. Further, as described above, a total of three types of prediction coefficients are required. In this pixel position relationship, line flicker is not emphasized.
[0030]
When performing field double speed conversion using another example of the pixel position relationship shown in FIG. ₁ , M ₂ , M _Three , M _Four The prediction accuracy differs for each of the four types of pixels. However, the variation is small compared to the example shown in FIG. 4, and the degree of variation is substantially the same as the example shown in FIG. M ₁ And m _Three Can be made common by using the symmetry in the time direction. Furthermore, m ₂ And m _Four Is created by using the symmetry of the space (vertical) direction. Person in charge Numbers can be shared. Therefore, two types of prediction coefficients are required in total for the four types of pixels to be created. In this case, line flicker is not emphasized.
[0031]
When performing field double speed conversion using another example of the pixel position relationship shown in FIG. ₁ , M ₂ , M _Three , M _Four The prediction accuracy differs for each of the four types of pixels. However, the variation is small compared to the example shown in FIG. 4, and the degree of variation is substantially the same as the example shown in FIG. M ₁ And m _Three Can be made common by using the symmetry in the time direction. Furthermore, m ₂ And m _Four Can be made common by using the symmetry in the spatial (vertical) direction. Therefore, two types of prediction coefficients are required in total for the four types of pixels to be created. In this case, line flicker is not emphasized.
[0032]
When performing field double speed conversion using still another example of the pixel positional relationship shown in FIG. ₁ , M ₂ , M _Three , M _Four The prediction accuracy is the same for each of the four types of pixels, and there is no variation depending on the pixel position of the prediction accuracy. Also, by using the symmetry in the time direction / the symmetry in the space (vertical) direction, m ₁ ~ M _Four It is possible to share all the prediction coefficients. Therefore, all of the four types of pixels to be created can be created under one type of prediction coefficient. In this case, line flicker may be emphasized.
[0033]
Considering the characteristics of each example of the pixel position relationship as described above, it is related to hardware restrictions such as memory capacity for holding the prediction coefficient, usage such as whether line flicker is allowed, etc. By selecting the most suitable pixel positional relationship according to conditions and the like, it is possible to improve the overall conversion performance.
[0034]
Next, learning in one embodiment of the present invention, that is, processing for determining a prediction coefficient will be described with reference to FIG. An image including all pixels in an image (referred to as a teacher image) having the same signal format as the image to be generated as a result of image information conversion is supplied via the input terminal 20. Since one embodiment of the present invention converts a 50 Hz image into a 100 Hz image, a 100 Hz progressive image is supplied to the vertical phase shift filter 21 and the delay circuit 28 via the input terminal 20. The vertical phase shift filter 21 performs processing according to which of the pixel positional relationships described above with reference to FIGS. 4 to 7 is used for learning (this will be described later).
[0035]
The output of the vertical phase shift filter 21 is supplied to the time thinning filter 22. The time thinning filter 22 generates a pixel at a 50 Hz interlaced pixel position by performing a process of smoothing the output of the vertical phase shift filter 21 in the time direction. In this way, the same signal format as that of the SD image (that is, a 50 Hz interlaced image) can be generated. Such an image is called a student image. The student image is supplied to the area cutout circuits 23 and 26.
[0036]
The area cutout circuit 23 performs the same processing as the area cutout circuit 2 in FIG. That is, the area cutout circuit 23 cuts out the same pixel area as that extracted by the area cutout circuit 2 from the supplied student image as a class tap, and supplies the cut out class tap data to the ADRC circuit 24. The ADRC circuit 24 performs the same processing as the ADRC circuit 3 in FIG. 1 on the supplied data. That is, by performing an operation as shown in the above equation (1), a requantization code in units of 2 bits is formed from data in units of 8 bits, for example. The formed requantization code is supplied to the class code generation circuit 25 as pattern compressed data.
[0037]
The class code generation circuit 25 performs the same processing as the class code generation circuit 3 in FIG. That is, the class code generation circuit 25 generates a class code by performing an operation as shown in the above equation (2) based on the output of the ADRC circuit 24. The output of the class code generation circuit 25 is supplied to a normal equation addition circuit 29. On the other hand, the area cutout circuit 26 performs the same processing as the area cutout circuit 6 in FIG. That is, the region cutout circuit 26 cuts out the same pixel region as that extracted by the region cutout circuit 6 from the supplied image signal as a prediction tap, and supplies the extracted prediction tap data to the delay circuit 27. The delay circuit 27 once holds the supplied data and then supplies it to the normal equation adding circuit 29.
[0038]
The delay circuit 28 once holds the supplied 100 Hz progressive image, and then supplies it to the normal equation adding circuit 29. The delay circuits 28 and 29 ensure that the normal equation addition circuit 29 is supplied with the output of the region cutout circuit 26 and the 100 Hz progressive image at a timing suitable for performing the arithmetic processing.
[0039]
The normal equation adding circuit 29 is based on the class tap data supplied from the delay circuit 27, the class code supplied from the delay circuit 25, and the 100 Hz progressive image supplied from the delay circuit 28. Calculate the data. Here, the normal equation adding circuit 29 extracts data constituting the teacher image from the 100 Hz progressive image, and performs arithmetic processing based on the extracted data. The calculated data is supplied to the prediction coefficient determination circuit 30. The prediction coefficient determination circuit 30 performs an operation for solving a normal equation based on the supplied data. The prediction coefficient calculated as a result of the calculation is supplied to the memory 31. Prior to performing the processing with the configuration described above with reference to FIG. 1, the storage contents of the memory 31 are loaded into the coefficient memory 5.
[0040]
A calculation process for calculating the prediction coefficient will be described. In the following description, the number of pixels cut out as a prediction tap is generalized to n. That is, x ₁ , X ₂ , ..., x _n Is a prediction tap, the pixel level y in the field double speed image is calculated according to the following equation (4).
[0041]
y = w ₁ X ₁ + W ₂ X ₂ + ... + w _n X _n (4)
Before learning, prediction coefficient w ₁ , ..., w _n Is an undetermined coefficient. Learning is performed for a plurality of image data for each class. When the number of types of image data is expressed as m, the following equation (5) is set from equation (4).
[0042]
y _k = W ₁ X _k1 + W ₂ X _k2 + ... + w _n X _kn (5)
(K = 1, 2,..., M)
If m> n, w ₁ , ..., w _n Is not uniquely determined, the element e of the error vector e _k Is defined by the following equation (6), and the prediction coefficient is determined so as to minimize the error vector e defined by equation (7). That is, the prediction coefficient is uniquely determined by a so-called least square method.
[0043]
e _k = Y _k -{W ₁ X _k1 + W ₂ X _k2 + ... + w _n X _kn } (6)
(K = 1, 2, ... m)
[0044]
[Expression 2]

[0045]
E in equation (7) ² As a practical calculation method for obtaining the prediction coefficient that minimizes ² Prediction coefficient w _i (i = 1, 2...) is partially differentiated (equation (8)), and each prediction coefficient w _i You just have to decide.
[0046]
[Equation 3]

[0047]
From Equation (8), each prediction coefficient w _i A specific procedure for determining the will be described. X as in equations (9) and (10) _ji , Y _i (8) can be written in the form of the determinant of equation (11).
[0048]
[Expression 4]

[0049]
[Equation 5]

[0050]
[Formula 6]

[0051]
Equation (11) is generally called a normal equation. The normal equation adding circuit 29 performs an operation as shown in the equations (9) and (10) based on the supplied data, so that X _ji , Y _i (I = 1, 2,..., N) are calculated. The prediction coefficient determination circuit 30 solves the normal equation (11) according to a general matrix solution method such as a sweep-out method, so that the prediction coefficient w _i (I = 1, 2,..., N) is calculated. As a result of the processing as described above, a prediction coefficient for estimating an estimated value statistically closest to the true value as field double speed data y for each class is calculated.
[0052]
In the present invention, learning as described above is selectively performed for each of a plurality of pixel positional relationships. In one embodiment of the present invention, a prediction coefficient corresponding to any of the four types of pixel positional relationships described above with reference to FIGS. 4 to 7 can be calculated. First, in the pixel position relationship as described above with reference to FIGS. 5 and 7, the pixel position in the vertical direction is shifted between the pixel in the input 50 Hz image and the pixel in the 100 Hz image to be created. When learning corresponding to such a positional relationship is performed, the inter-vertical phase shift filter 21 performs a phase shift in the vertical direction with respect to the supplied 100 Hz progressive image.
[0053]
On the other hand, in the pixel positional relationship as described above with reference to FIGS. 4 and 6, the pixel position in the vertical direction of the input pixel in the 50 Hz image and the pixel in the 100 Hz image to be created are the same. . When performing learning corresponding to such a positional relationship, the vertical thinning filter 21 does not perform phase shift and outputs the supplied 100 Hz progressive image as it is.
[0054]
As described above, among the plurality of types of pixel positional relationships between the 50 Hz image and the 100 Hz image, which are objects and results of the image information conversion, higher as a 100 Hz image to be generated as a result of the image information conversion. It is possible to perform learning such that a quality image is generated.
[0055]
Also, learning corresponding to the most suitable pixel positional relationship in relation to hardware restrictions such as memory capacity for holding prediction coefficients, conditions such as whether or not line flicker is allowed, etc. Can be selectively performed. As a result, it is possible to perform image information conversion such as field double speed conversion by applying a class classification adaptation process that presupposes a more preferable pixel position relationship, and thus it is possible to improve the overall conversion performance.
[0056]
One embodiment of the present invention described above performs field double speed conversion for converting a 50 Hz image into a 100 Hz image. On the other hand, at the time of learning, for example, a teacher image or a student image is generated by performing vertical phase shift or the like as necessary, thereby converting an interlaced image into a progressive image, or an SD image as an HD ( (High Definition) The present invention can also be applied to image information conversion such as conversion to an image. That is, the present invention is effective in general image information conversion for converting a digital image into a digital image having a larger number of pixels.
[0057]
In addition, although one embodiment of the present invention performs learning between a 100 Hz progressive image and a 50 Hz interlaced image, the combination of images related to learning is not limited to this. For example, learning may be performed between a 50 Hz progressive image and a 25 Hz interlaced image, or learning may be performed between a 60 Hz progressive image and a 30 Hz interlaced image.
[0058]
In the embodiment of the present invention described above, ADRC is performed to pattern the spatial waveform of an input image such as a 50 Hz image with a small number of bits. On the other hand, if it is an information compression method that contributes to expressing an image pattern in space-time with a small number of classes, a method other than ADRC is used In You may do it. For example, a method such as DPCM (Differential Pulse Code Modulation) or VQ (Vector Quantization) may be used.
[0059]
The present invention is not limited to the above-described embodiment of the present invention, and various modifications and applications can be made without departing from the gist of the present invention.
[0060]
【The invention's effect】
According to the present invention, a prediction coefficient determined by selecting any combination from among a plurality of combinations of student images and teacher images and performing learning under the selected combination is used. Predictive calculations can be performed.
[0061]
Therefore, it is possible to perform learning such that a higher quality image is generated as a result of the image information conversion.
[0062]
Also, learning corresponding to the most suitable pixel positional relationship in relation to hardware restrictions such as memory capacity for holding prediction coefficients, conditions such as whether or not line flicker is allowed, etc. Can be performed.
[0063]
In addition, since the number of types of prediction coefficients varies depending on the combination of the student image and the teacher image, the number of types of prediction coefficients varies, so the amount of data to be stored as the prediction coefficient is selected. It depends on the combination of the student image and the teacher image. For this reason, by applying the present invention, it is possible to perform image information conversion that is more suitable for conditions such as the capacity of the memory for prediction coefficients allowed in the specifications of the apparatus.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating an example of a configuration related to image information conversion according to an embodiment of the present invention.
FIG. 2 is a schematic diagram illustrating an example of a tap structure of a class tap.
FIG. 3 is a schematic diagram illustrating an example of a tap structure of a prediction tap.
FIG. 4 is a schematic diagram for explaining a pixel position relationship in image information conversion according to the previous proposal of the applicant of the present application.
FIG. 5 is a schematic diagram for explaining an example of a pixel position relationship that can be used in an embodiment of the present invention;
FIG. 6 is a schematic diagram for explaining another example of the pixel positional relationship that can be used in one embodiment of the present invention.
FIG. 7 is a schematic diagram for explaining yet another example of a pixel positional relationship that can be used in an embodiment of the present invention.
FIG. 8 is a block diagram for explaining learning in an embodiment of the present invention.
[Explanation of symbols]
21 ... Vertical phase shift filter

Claims (4)

In an image information conversion apparatus for forming an interlace scanning output digital image signal having a number of fields larger than the number of fields of the input digital image signal by generating a new field between fields of the interlace scanning input digital image signal.
First image cutout means for cutting out a plurality of image data around a target pixel to be processed from the input digital image signal;
The level distribution of each of the plurality of image data cut out by the first image cutout means is detected, the class to which the target pixel belongs is determined from the detected level distribution, and class detection information representing the determined class is output. Class detection means to
Second image cutout means for cutting out a plurality of image data around the pixel of interest from the input digital image signal;
Prediction corresponding to each of the classes, a prediction coefficient for estimating the output digital image signal is stored, and a prediction corresponding to the class detection information from the class detection means is stored out of the stored prediction coefficients Coefficient storage means for outputting a coefficient;
An operation for performing a product-sum operation on the plurality of image data cut out by the second image cut-out means and a prediction coefficient supplied from the coefficient storage means to generate a predicted value of the image data of the output digital image signal Processing means,
The prediction coefficient is obtained by learning in advance for each class and stored in the coefficient storage means.
The above learning
A first positional relationship that includes a field that coincides with the field of the input digital image signal in the time direction with respect to the pixel position of the input digital image signal and that has a pixel position in the vertical direction that coincides with the input digital image signal. A second positional relationship that includes an output digital image signal and a field that coincides with the field of the input digital image signal in the time direction, and that the pixel position is symmetrical in the vertical direction about the pixel position of the input digital image signal. An output digital image signal, and an output digital image signal having a third positional relationship in which the field is symmetric in the time direction around the field of the input digital image signal and the pixel position in the vertical direction coincides with the input digital image signal; The field centered on the input digital image signal field De becomes symmetrical in the time direction, and consists of a output digital image signal of the fourth positional relationship pixel position as the center pixel position of the input digital image signal is symmetrical in the vertical direction,
An output digital image signal having one positional relationship is selected from the output digital image signals having the first, second, third, and fourth positional relationships, and the same signal format as the selected output digital image signal Converting the pixel of the teacher image signal into the pixel position of the input digital image signal ,
Generating a digital image signal in which the number of fields per unit time of the converted output digital image signal is converted to be the same as the input digital image signal;
By the product-sum operation, and a true value of the pixel at the corresponding position of at, the pixel values generated and the teacher image signal to generate a pixel value of the selected output digital image signal from a digital image signal generated as above An image information conversion apparatus which is a process for obtaining the prediction coefficient so as to minimize the error. The arithmetic processing means is:
2. The image information conversion apparatus according to claim 1, wherein the output digital image signal is generated so that the number of fields per unit time is twice that of the input digital image signal. In the above learning,
2. The image information conversion apparatus according to claim 1, further comprising vertical phase shift means for generating a vertical phase shift with respect to a pixel of the output digital image signal and converting the pixel to a pixel position of the input digital image signal . In an image information conversion method for forming an interlace scanning output digital image signal having a number of fields larger than the number of fields of the input digital image signal by generating a new field between fields of the interlace scanning input digital image signal.
A first image cutout step of cutting out a plurality of pieces of image data around a target pixel to be processed from the input digital image signal;
The level distribution of each of the plurality of image data cut out by the first image cut-out step is detected, the class to which the target pixel belongs is determined from the detected level distribution, and class detection information expressing the determined class is output. Class detection step to
A second image cutout step of cutting out a plurality of image data around the pixel of interest from the input digital image signal;
Prediction corresponding to each of the classes, a prediction coefficient for estimating the output digital image signal is stored, and a prediction corresponding to the class detection information from the class detection step is stored out of the stored prediction coefficients A coefficient storage step for outputting a coefficient;
An operation for performing a product-sum operation on the plurality of image data cut out by the second image cut-out step and a prediction coefficient supplied from the coefficient storage step to generate a predicted value of the image data of the output digital image signal Processing steps,
The prediction coefficient is obtained by learning in advance for each class and stored,
The above learning
A first positional relationship that includes a field that coincides with the field of the input digital image signal in the time direction with respect to the pixel position of the input digital image signal and that has a pixel position in the vertical direction that coincides with the input digital image signal. A second positional relationship that includes an output digital image signal and a field that coincides with the field of the input digital image signal in the time direction, and that the pixel position is symmetrical in the vertical direction about the pixel position of the input digital image signal. An output digital image signal, and an output digital image signal having a third positional relationship in which the field is symmetric in the time direction around the field of the input digital image signal and the pixel position in the vertical direction coincides with the input digital image signal; The field centered on the input digital image signal field De becomes symmetrical in the time direction, and consists of a output digital image signal of the fourth positional relationship pixel position as the center pixel position of the input digital image signal is symmetrical in the vertical direction,
An output digital image signal having one positional relationship is selected from the output digital image signals having the first, second, third, and fourth positional relationships, and the same signal format as the selected output digital image signal Converting the pixel of the teacher image signal into the pixel position of the input digital image signal ,
Generating a digital image signal in which the number of fields per unit time of the converted output digital image signal is converted to be the same as the input digital image signal;
By the product-sum operation, and a true value of the pixel at the corresponding position of at, the pixel values generated and the teacher image signal to generate a pixel value of the selected output digital image signal from a digital image signal generated as above An image information conversion method, which is a process for obtaining the prediction coefficient so as to minimize the error.

Images