CN103617431B - Maximum average entropy-based scale-invariant feature transform (SIFT) descriptor binaryzation and similarity matching method - Google Patents

Abstract

The invention relates to a maximum average entropy-based scale-invariant feature transform (SIFT) descriptor binaryzation and similarity matching method and belongs to the image matching field. Scale-invariant feature transform (SIFT) operators have strong matching ability, while, the scale-invariant feature transform (SIFT) operators will bring a huge amount of data, and therefore, binaryzation should be performed on the scale-invariant feature transform (SIFT) operators, however, if unified binaryzation is performed on all the operators, data redundancy or information loss will be brought about. The maximum average entropy-based scale-invariant feature transform (SIFT) descriptor binaryzation and similarity matching method of the invention comprises the following steps that: binaryzation is performed on the scale-invariant feature transform (SIFT) operators; average entropy calculation is performed on each layer of binaryzation results, such that different numbers of binaryzation layers are selected adaptively; new binaryzation descriptors are provided; the Hamming distance is utilized to replace the Euclidean distance so as to calculate the distance between two descriptors; and the distance between the two descriptors is compared with a set threshold value. With the maximum average entropy-based scale-invariant feature transform (SIFT) descriptor binaryzation and similarity matching method of the invention adopted, information of original features is reserved; the amount of data storage can be greatly reduced; computational complexity can be reduced; a requirement for a real-time property can be realized better; the matching results equivalent to original scale-invariant feature transform (SIFT) descriptors can be obtained and are far superior to results of matching by using the unified binaryzation.

Description

SIFT Descriptor Binarization and Similarity Matching Method Based on Maximum Bit Average Entropy

technical field

The invention relates to the technical field of image matching, in particular to a SIFT descriptor binarization method based on maximum bit average entropy and a similarity matching scheme thereof.

Background technique

With the development of modern computer technology, face recognition technology has been widely used in security authentication, human-computer communication, public security system, etc., and also plays a great role in video conferencing, file management, medical treatment, etc. After the 911 terrorist attacks in the United States and the leakage of network CSDN user information, biometric technology has attracted more attention, and the recognition of facial biometrics has always been a hot spot in the field of biometrics. However, in practical applications, face recognition is often affected by many factors. When the face posture changes, the expression changes, the external light changes, and the face is blocked (wearing a scarf, sunglasses, etc.) ) and other situations, the performance of face recognition will drop a lot, which restricts the practical application of face recognition. Therefore, many researchers are devoted to the research of face recognition methods, and various face recognition methods emerge in endlessly.

The SIFT (Scale-Invariant Feature Transform) feature matching algorithm is a hot spot and difficulty in the research of feature point matching algorithms at home and abroad. It is a local feature descriptor proposed by David G. Lowe of Canada in 1999, and was developed in 2004 On the original basis, it has been further developed and improved. The SIFT descriptor is a local descriptor of an image, which is invariant to scale, rotation, and translation, and has certain robustness to illumination changes, affine transformations, and 3D projection transformations. In Mikolajczyk's invariance comparison experiments on ten kinds of local descriptors including SIFT descriptors, SIFT and its extended algorithm have been proved to have the strongest robustness among similar descriptors. The SIFT descriptor has strong matching ability, and has strong invariance to most image transformations. It is especially suitable for dealing with the matching problems when translation, rotation, and affine transformation occur between two images. Its stable feature matching ability can even be used for Application of images taken from any angle. SIFT features also have good uniqueness, which is suitable for relatively fast and accurate matching in massive feature databases. However, the amount of data to represent face images with SIFT is huge. Generally, a face image has 3000 SIFT points, each SIFT consists of 128 descriptors, each descriptor is represented by 8 bits, and the total data volume is 3072000 bits. Wang et al. proposed the idea of binarizing SIFT descriptors. In the present invention, when performing binarization on SIFT descriptors, the number of layers of binarization is determined with the maximum bit average entropy as a criterion to form a new binarization descriptor, and then the Hamming distance is used to replace the Euclidean distance to calculate two The distance between each descriptor is compared with the set threshold, and finally the matching result is obtained. In this way, the amount of data is greatly reduced, the complexity of calculation is reduced, and the requirement of real-time performance can be better realized. And the binarized descriptor can obtain the matching result that is basically the same as the original SIFT descriptor when performing the matching operation, which is far better than the result of uniformly binarizing the SIFT descriptor for matching. The invention proposes a SIFT descriptor binarization method and a similarity matching scheme based on maximum bit average entropy.

Contents of the invention

The present invention provides a SIFT descriptor binarization method and a similarity matching scheme based on the maximum bit average entropy criterion. The method can use the maximum bit average entropy to effectively determine the number of layers for binarization to form a new binary Value descriptors, and then use Hamming distance instead of Euclidean distance to calculate the distance between two descriptors, compare it with the set threshold, and finally get the matching result. A large amount of original SIFT descriptor information can be retained, the amount of data is reduced to a large extent, the complexity of calculation is reduced, and the requirement of real-time performance can be better realized. And it can guarantee that the matching result is basically the same as the matching result obtained by using the original SIFT descriptor, which is far better than the result of uniform binarization of the SIFT descriptor for matching.

To binarize the SIFT descriptor, if only one layer of binarization is performed, a lot of information will be lost, and the amount of information retaining the original features will be very small. When performing multi-layer binarization, four binarization strategies are included: (1) After the first layer of binarization is completed, 0 and 1 are obtained, and then the 0 part is always binarized. (2) After the first layer of binarization is completed, 0 and 1 are obtained, and the 0 part is selected for binarization to obtain 0 and 1, and then the 1 part is always binarized. (3) After the first layer of binarization is completed, 0 and 1 are obtained, and the 1 part is selected for binarization to obtain 0 and 1, and then the 0 part is always binarized. (4) After the first layer of binarization is done, 0 and 1 are obtained, and then the 1 part is always binarized. Because the SIFT descriptor contains many 0s, for each layer of binarization results, the strategy information entropy of binarizing 1 is always the largest, so the fourth strategy is adopted for multi-layer binarization. Entropy can be used to measure the amount of information. As the number of binarization layers of the SIFT descriptor increases, the information entropy will increase accordingly. However, it is accompanied by a great increase in the amount of data. Moreover, if all extracted SIFT descriptors are binarized with a uniform number of layers, the following two situations will be included: (1) Binarization itself does not need to be performed to a specified number of layers, and the average bit entropy has reached the maximum. For example, the average bit entropy after two layers of binarization is the largest, and only the results of the two layers need to be retained. If three or more layers are used, data redundancy will be increased, and redundant data may cause matching errors; (2) After the binarization of the specified number of layers is completed, the average bit entropy has not yet reached the maximum, This will lead to the fact that the binarized descriptor cannot fully retain the information carried by the original SIFT descriptor, which will lead to an increase in the degree of false matching. However, considering the amount of data, up to four layers of binarization results are retained. The present invention proposes a binarized SIFT descriptor method and a similarity matching scheme based on maximum bit average entropy. By calculating bit average entropy for each descriptor, different binarization layers are adaptively selected to form a new The binary descriptor, and then use the Hamming distance instead of the Euclidean distance to calculate the distance between the two descriptors, compare it with the set threshold, and finally get the matching result. The algorithm proposed by the invention retains the original feature information to a large extent, greatly reduces the amount of data storage, reduces the complexity of calculation, and can better realize the real-time requirement. And the matching result equivalent to the original SIFT descriptor can be obtained, which is far better than the result of uniform binarization of the SIFT descriptor for matching. Therefore, the present invention has certain application value and significance.

In order to realize the above-mentioned problem, the present invention proposes a kind of method and similarity matching scheme based on the SIFT descriptor binarization of maximum bit average entropy, and this method specifically comprises:

A. In the binarization stage, for each face image, first extract the SIFT feature descriptor, and then perform multi-layer binarization on the SIFT descriptor to obtain the information entropy of each level, according to the appearance of 0 and 1 in each layer The probability and the total number of reserved bits are used to obtain the average bit entropy, and then the number of binarization layers corresponding to the maximum bit average entropy is found, and the binarization results of these layers are retained to form a new binary feature descriptor.

B. In the matching stage, for any two face images, after extracting their new binary feature descriptors, the Hamming distance is used instead of the Euclidean distance to calculate the distance between the two descriptors, which is compared with the set threshold For comparison, if the Hamming distance is less than or equal to the set threshold, the matching is considered successful; otherwise, the matching is considered unsuccessful. For the extracted binary feature descriptors of different levels, different thresholds need to be selected.

Described step A specifically comprises:

A1, for each face image, first extract the SIFT feature descriptor;

A2, perform binarization for each SIFT descriptor of each face image;

A3, count the number of 0 and 1, respectively n ₁₀ and n ₁₁ ;

A4. The descriptors corresponding to those bytes marked as 1 continue to be binarized, and the numbers of 0 and 1 are counted at this time, which are n ₂₀ and n ₂₁ respectively. Continue the above binarization process to obtain n ₃₀ in turn and n ₃₁ , n ₄₀ and n ₄₁

A5. Calculate the information entropy of binarization of the first layer, the second layer, the third layer, and the fourth layer respectively;

A6. According to the total number of bits retained after binarization, the average bit entropy can be obtained by dividing the information entropy by the number of bits;

A7. Obtain the number of layers corresponding to the maximum bit average entropy, retain the binarization results of the corresponding layers, and form a new binarization descriptor.

Described step B specifically comprises:

B1. For any two face images, extract their new binary descriptors;

B2. For every two descriptors, calculate the Hamming distance dis _H ;

B3. To extract binary descriptors of different levels, the selected threshold T is also different. The thresholds for one-layer, two-layer, three-layer and four-layer binarization are: T ₁ , T ₂ , T ₃ , T ₄

B4. If dis _H is less than or equal to the threshold, it is considered that the matching is successful; otherwise, it is considered that the matching is unsuccessful. Compared with the prior art, the present invention has the following obvious advantages and beneficial effects:

(1) The present invention carries out binarization to SIFT descriptor, is to determine binarization layer number according to each descriptor self-adaptively according to the principle of maximum bit average entropy, compared with carrying out unified binarization, not only reduces Data redundancy, and to a large extent retain the information carried by the original SIFT descriptor.

(2) The present invention self-adaptively obtains a new binary descriptor, and performs matching through Hamming distance, which is simple and quick to calculate, reduces complexity, and can better meet real-time requirements.

(3) The present invention matches through the binarized SIFT descriptor obtained by the maximum bit average entropy. Through experimental analysis, the matching result is equal to the matching result of the original SIFT descriptor, which is far superior to unifying the SIFT descriptor. The result obtained by value-based matching.

Description of drawings:

Fig. 1 is an overall flow chart of the technical solution.

Figure 2(a) is binarization strategy 1 (left left)

Figure 2(b) is binarization strategy 1 (left and right)

Figure 2(c) is binarization strategy 1 (right left)

Figure 2(d) is binarization strategy 1 (right right)

Figure 3 is a comparison chart of similarity matching results.

detailed description:

The overall flow of the technical solution of the present invention is shown in Figure 1 of the description. Our method greatly reduces the amount of data storage, reduces the complexity of calculation, and can better meet the requirements of real-time performance. And the matching result equivalent to the original SIFT descriptor can be obtained, which is far better than the result of uniform binarization of the SIFT descriptor for matching.

A. For each face image, first extract the SIFT feature descriptor, and then binarize the SIFT descriptor to obtain the information entropy of each level, according to the probability of occurrence of 0 and 1 in each layer and the total number of bits reserved , to obtain the bit average entropy, and then find out the number of binarization layers corresponding to the maximum bit average entropy, keep the binarization results of these layers, and finally form a new binarization descriptor. Specific steps include:

A1. For each face image, first extract the SIFT feature descriptor

D=(f ₁ ,f ₂ ,…,f ₁₂₈ ) ^T ∈ R ¹²⁸

A2. Binarize each SIFT descriptor of each face image

is the median value of the vector D

A3, count the number of 0 and 1, respectively n ₁₀ and n ₁₁ ;

A4. The descriptors corresponding to those bytes marked as 1 continue to be binarized, and the numbers of 0 and 1 are counted at this time, which are n ₂₀ and n ₂₁ respectively. Continue the above binarization process to obtain n ₃₀ in turn and n ₃₁ , n ₄₀ and n ₄₁

A5. Calculate the information entropy after binarization of each layer

The information entropy for one layer of binarization is:

H ₁ =-P(n ₁₀ )log ₂ P(n ₁₀ )-P(n ₁₁ )log ₂ P(n ₁₁ )

The information entropy for two-layer binarization is:

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}<span\; class="notranslate">\; )</span>$

The information entropy for three-layer binarization is:

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}<span\; class="notranslate">\; )</span>$

The information entropy of four-layer binarization is:

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}<span\; class="notranslate">\; )</span>$

A6. Calculate the number of bits retained after each layer of binarization

The total number of bytes reserved after one layer of binarization is:

N ₁ =n ₁₀ +n ₁₁

The total number of bytes reserved after the second layer binarization is:

${\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}$

The total number of bytes reserved after three layers of binarization is:

${\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}$

The total number of bytes reserved after four layers of binarization is:

${\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}$

A7. The average bit entropy can be obtained by dividing the information entropy by the number of bits

The average bit entropy after one layer of binarization is:

${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; h</span>}}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}<span\; class="notranslate">\; )</span>$

The average bit entropy after two-layer binarization is:

${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; h</span>}}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}<span\; class="notranslate">\; )</span>$

The average bit entropy after three layers of binarization is:

${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; h</span>}}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}<span\; class="notranslate">\; )</span>$

The average bit entropy after four layers of binarization is:

${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; h</span>}}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 41</span>}}<span\; class="notranslate">\; )</span>$

A8. Obtain the number of layers corresponding to the maximum bit average entropy, keep the binarization results of the corresponding layers, and form a new binarization descriptor.

B. In the matching stage, for any two face images, after extracting their new binary feature descriptors, the Hamming distance is used instead of the Euclidean distance to calculate the distance between the two descriptors, and the threshold is set For comparison, if the Hamming distance is less than or equal to the set threshold, the matching is considered successful; otherwise, the matching is considered unsuccessful. For the extracted binary feature descriptors of different levels, different thresholds need to be selected. Specific steps include:

B1. For any two face images, extract their new binary descriptors;

B2. For every two descriptors x=[b ₁₁ ,b ₁₂ ,b ₁₃ ,…] and y=[b ₂₁ ,b ₂₂ ,b ₂₃ ,…], calculate their Hamming distance

${\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&CirclePlus;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}$

B3. Comparing the obtained Hamming distance with the threshold value, the selected threshold value is also different for the binary descriptors of different levels, and the threshold value of the i-layer binarization is set to T _i (i=1,2,3 ,4, T ₁ =17; T ₂ =28; T ₃ =34; T ₄ =39)

$\mathrm{<span\; class="notranslate">\; S</span>}\mathrm{<span\; class="notranslate">\; G</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\end{array}\right.$

B4. If SGN(Dis _H )=1, it is considered that the matching is successful; otherwise, it is considered that the matching is not successful.

The matching results are shown in Figure 3 of the specification, (a) is the matching result of the original SIFT descriptor, and the number of matches is 1242; (b) is the matching result after unified binarization, and many wrong matches have been generated. (c) is the matching result of the method of the present invention, and the matching number is 1161. The matching result of the method of the present invention is equal to the matching result of the original SIFT descriptor, which is far superior to the result obtained by uniformly binarizing the SIFT descriptor for matching.

Claims (1)

1. a kind of sub- binarization method and similarity mode scheme are described based on the sift of dominant bit mean entropy, walk including following Rapid:

A, binaryzation stage, for each width facial image, extract sift Feature Descriptor first, then son is described to sift and enter Row multilamellar binaryzation, tries to achieve comentropy at all levels, the total bit number of the probability being occurred according to each layer 0 and 1 and reservation, tries to achieve Position mean entropy, then finds out the dominant bit mean entropy corresponding binaryzation number of plies, retains this which floor binaryzation result, forms new two Value Feature Descriptor；

Described step a specifically includes:

A1, for each width facial image, extract sift Feature Descriptor first

D=(f₁,f₂,…,f₁₂₈)^t∈r¹²⁸

A2, each sift for each width facial image describe son and carry out binaryzation

Intermediate value for vectorial d

A3, the number of statistics 0 and 1, respectively n₁₀And n₁₁；

A4, be labeled as 1 those bytes corresponding description, proceed binaryzation, statistics now 0 and 1 number, respectively n₂₀And n₂₁, proceed above-mentioned binarization, obtain n successively₃₀And n₃₁, n₄₀And n₄₁；

A5, calculating do the comentropy after each layer binaryzation

The comentropy doing one layer of binaryzation is:

h₁=-p (n₁₀)log₂p(n₁₀)-p(n₁₁)log₂p(n₁₁)

The comentropy doing two layers of binaryzation is:

$h_2=-\underset{i=1}{\overset{2}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(n_{i0}\right)-p\left(n_{21}\right){\log }_{2}p\left(n_{21}\right)$

The comentropy doing three layers of binaryzation is:

$h_3=-\underset{i=1}{\overset{3}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(h_0\right)-p\left(n_{31}\right){\log }_{2}p\left(n_{31}\right)$

The comentropy doing four layers of binaryzation is:

$h_4=-\underset{i=1}{\overset{4}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(n_{i0}\right)-p\left(n_{41}\right){\log }_{2}p\left(n_{41}\right)$

The bit number that a6, calculating retain after each layer binaryzation

The total bytes retaining after doing one layer of binaryzation are:

n₁=n₁₀+n₁₁

The total bytes retaining after doing two layers of binaryzation are:

$n_2=\underset{i=1}{\overset{2}{\&sigma;}}{n}_{i0}+n_{21}$

The total bytes retaining after doing three layers of binaryzation are:

$n_3=\underset{i=1}{\overset{3}{\&sigma;}}{n}_{i0}+n_{31}$

The total bytes retaining after doing four layers of binaryzation are:

$n_4=\underset{i=1}{\overset{4}{\&sigma;}}{n}_{i0}+n_{41}$

A7, position mean entropy can be obtained divided by digit by comentropy

Doing the position mean entropy after one layer of binaryzation is:

${\stackrel{\&overbar;}{h}}_1=h_1/n_1=\&lsqb;-p\left(n_{10}\right){\log }_{2}p\left(n_{10}\right)-p\left(n_{11}\right){\log }_{2}p\left(n_{11}\right)\&rsqb;/(n_{10}+n_{11})$

Doing the position mean entropy after two layers of binaryzation is:

${\stackrel{\&overbar;}{h}}_2=h_2/n_2=\&lsqb;-\underset{i=1}{\overset{2}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(n_{i0}\right)-p\left(n_{21}\right){\log }_{2}p\left(n_{21}\right)\&rsqb;/(\underset{i=1}{\overset{2}{\&sigma;}}{n}_{i0}+n_{21})$

Doing the position mean entropy after three layers of binaryzation is:

${\stackrel{\&overbar;}{h}}_3=h_3/n_3=\&lsqb;-\underset{i=1}{\overset{3}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(n_{i0}\right)-p\left(n_{31}\right){\log }_{2}p\left(n_{31}\right)\&rsqb;/(\underset{i=1}{\overset{3}{\&sigma;}}{n}_{i0}+n_{31})$

Doing the position mean entropy after four layers of binaryzation is:

${\stackrel{\&overbar;}{h}}_4=h_4/n_4=\&lsqb;-\underset{i=1}{\overset{4}{\&sigma;}}p\left(n_{i0}\right){\log }_{2}p\left(n_{i0}\right)-p\left(n_{41}\right){\log }_{2}p\left(n_{41}\right)\&rsqb;/(\underset{i=1}{\overset{4}{\&sigma;}}{n}_{i0}+n_{41})$

A8, the number of plies tried to achieve corresponding to dominant bit mean entropy retain the binaryzation result of the corresponding number of plies, form new binaryzation and retouch State son；

B, matching stage, for any two width facial images, after extracting its new binaryzation Feature Descriptor respectively, Ran Houyong Hamming distance replaces Euclidean distance to calculate the distance between two description, it is compared with given threshold, if Hamming Distance is less than or equal to given threshold then it is assumed that the match is successful；Otherwise then it is assumed that coupling is unsuccessful；For the different levels extracted Binaryzation Feature Descriptor, need to select different threshold values；

Described step b specifically includes:

B1, for any two width facial images, extract its new binaryzation description；

B2, sub- x=[b is described for each two₁₁,b₁₂,b₁₃...] and y=[b₂₁,b₂₂,b₂₃...], calculate its Hamming distance

${\mathrm{dis}}_h(x,y)=x\&circleplus;y$

B3, the Hamming distance obtaining and threshold value are compared, the binaryzation for different levels describes son, the threshold value of selection Differ, set the threshold value of i layer binaryzation as t_i, wherein i=1,2,3,4, t₁=17；t₂=28；t₃=34；t₄=39；

$sgn\left({\mathrm{dis}}_h\right)=\left\{\begin{array}{c}1,di{s}_h\&le;t_i\\ 0,di{s}_h>t_i\end{array}\right.$

If b4 is sgn (dis_h)=1 is then it is assumed that the match is successful；Otherwise it is assumed that coupling is unsuccessful.