JP3198705B2 - Image feature extraction method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image feature extraction method and apparatus for extracting pixels having image features from quantized continuous multi-tone image data.

[0002]

2. Description of the Related Art In a conventional image processing apparatus,
It is common to extract image feature information for each individual image process such as edge enhancement and smoothing, and to perform processing appropriate for each result based on the result. For example, when performing binarization or edge enhancement processing, it is necessary to determine whether the pixels constituting the image are part of a character, a part of a halftone image, or a part of a photograph. The feature information is extracted from the image, and the result is subjected to a binarization process or an edge enhancement process. In various recognition processes such as region extraction and region division (segmentation),
A process for feature extraction is performed in the first half of the process. In this case, as a method of extracting image features, a method of selecting the same feature for each region of the image,
There is a method of selecting a feature for each pixel, and another approach is to convert a binary image such as a halftone dot into a multi-valued image and apply the same processing to both ( For example, JP-A-3-283765).

[0003]

By the way, in a conventional image processing apparatus, several kinds of individual image processing units are incorporated in the same apparatus, but even if similar image features are used. There is a technical problem that the device configuration is complicated because features are separately extracted, used, or stored. Furthermore, the conventional binarization and edge enhancement processing merely changes the entire processing depending on whether the image is a photograph, a halftone dot, or a character, and adds an appropriate processing to an appropriate place at an appropriate depth. Therefore, there is a technical problem that individual image processing accuracy such as edge enhancement processing cannot be sufficiently improved.

SUMMARY OF THE INVENTION The present invention has been made to solve the above technical problems, and has a compact system as a whole by effectively utilizing general-purpose image features extracted from multi-tone image data. In addition, the present invention provides an image feature extraction method and an apparatus for extracting an image feature with a fine classification, and realizing accurate individual image processing while referring to the extracted image feature.

[0005]

That is, the present invention provides:
How to extract input features to human visual characteristics when extracting image features from quantized multi-tone image data
, And based on this prediction result,
Through pre-processing of data processing based on visual characteristics
Create multi-tone image data to extract characteristic pixels,
After, the pixels of each scan line, after extracting the feature pixel commonly available in a plurality of individual image processing means, for reference the nature of the extracted feature pixels are a plurality of individual image processing means may share The data is written in the corresponding pixel position of the buffer.

[0006] Such a method invention apparatus invention embodying, as shown in FIG. 1, an image characteristic extraction apparatus for extracting image features from the multi-gradation image data quantized, picture
Pre-processing means for multi-tone image data to extract image features
To predict how the input image will appear in human visual characteristics.
Image predicting means (5) for measuring, and the image predicting means (5)
Input image based on the human visual characteristics
Data shaping means (6) for performing data processing according to
So, from the pixels of each scan line, wherein the pixel extracting means 1 for extracting a feature pixel TD which contains the edge feature pixel TDE which distinguishes between and rise and fall in response to a large edge pixel portion of at least the concentration change, the And a reference buffer (2) in which the characteristics of the characteristic pixels TD extracted by the characteristic pixel extracting means (1) are written for each corresponding pixel and provided so as to be shared by a plurality of individual image processing means (7). It is assumed that. In another apparatus invention, the characteristic pixel extracting means 1 shown in FIG. 1 extracts a characteristic pixel TD including a density gradient characteristic pixel TDN representing a density gradient of a pixel portion having a density gradient, or at least a density pixel. One that extracts a feature pixel TD that includes an edge feature pixel TDE corresponding to an edge pixel portion having a large change and a density gradient feature pixel TDN representing a density gradient of a pixel portion having a density gradient is used.

In such technical means, the input image itself may be used as the multi-tone image data from which the image feature is to be extracted, but from the viewpoint of further improving the extraction accuracy of the image feature, the input image can be used. preferably one subjected to correction processing based on the human visual characteristics for. Here, the correction processing based on the visual characteristics of a human means that the input image data is converted into a space in which a human actually perceives the input image (including an illusion).
The correction algorithm can be appropriately selected.

In order to perform a correction process based on human visual characteristics , as a preprocessing means for multi-tone image data to be subjected to image feature extraction, it is necessary to determine how an input image appears in human visual characteristics. What is necessary is just to provide the image prediction means 5 for predicting, and the data shaping means 6 which performs the data processing based on the human visual characteristic with respect to the input image based on the prediction result of this image prediction means 5. At this time, if the image predicting means 5 predicts how an image is projected to human vision, the algorithm of the prediction can be appropriately selected. May be appropriately selected as long as the data is shaped in accordance with the human visual characteristics based on the prediction result of the image predicting means 5. However, in order to maintain good data shaping, the input image must be equalized. It is preferable to adopt a method of mapping in a color space, and particularly from the viewpoint of improving data processing, it is preferable to handle data in a polar coordinate display in a uniform color space.

The following is a background where it is considered preferable to perform image feature extraction on multi-tone image data that has been subjected to correction processing based on human visual characteristics. That is, for example, regarding image processing of an image reproducing apparatus, individual image processing such as edge enhancement processing is one type of processing for converting an input image into an image that can be actually seen by humans in view of human visual characteristics. Can be regarded as one. Thereafter, the input image subjected to various processes is converted again according to the characteristics of the respective image output devices such as binarization and output. Therefore, as an image processing function to be required of an image reproducing apparatus, a conversion unit (hereinafter referred to as HVS (Human Visual Sy
stem), and an encoding unit that performs a process such as binarization in accordance with the characteristics of the image output device. At this time, correct HVS conversion cannot be performed unless proper consideration is given to how an image can be seen through human vision. Even if it does, the reproducibility of the image cannot be improved. In addition, HVS
For example, Olivier D. Faugeras's “DigitalColor Image Processing Within the Framewo
rk of a Human Visual Model "IEEEtransactions, ASSP27
no.4, pp380-393 and Hannu J. Saarelma and Pirkko T. Oi
ttinen's “Digital Halftones and Structure Visibili
ty "Journal of Imaging Technology17, pp228-232,1991
And the like. Therefore, if the input image is first subjected to HVS conversion, it becomes possible to understand how photographs, characters, and halftone dots are seen by human vision.
It is considered that individual image processing such as appropriate edge enhancement processing or the like can be performed at an appropriate location with an appropriate strength.

As described above, the characteristic pixel TD extracted by the characteristic pixel extracting means 1 is, as described above, an edge characteristic pixel TDE in which rising and falling are distinguished, a density gradient characteristic pixel TDN, or edge feature pixel TDE may be contained the (rising and the falling edges including those not differentiated) and the concentration gradient feature pixel TDN, but the one corresponding to an image feature that is required in the individual image processing Of course, it may be included. Also,
The extraction algorithm of the characteristic pixel TD by the characteristic pixel extraction means 1 may be appropriately selected, but from the viewpoint of improving the extraction accuracy of the characteristic pixel TD, a plurality of pixels located around the pixel of interest are referred to. It is preferable to adopt a method of extracting the characteristic pixel TD of the target pixel.

The reference buffer 2 only needs to have a minimum required capacity in accordance with the method of extracting the characteristic pixels TD by the characteristic pixel extracting means 1 and the number of individual image processing means. In addition, the expression format of the characteristic pixel TD to be written in the reference buffer 2 may be appropriately selected as long as it is predetermined.

Further, from the viewpoint of further improving the extraction accuracy of the characteristic pixel TD, it is determined whether or not the characteristic pixel TD written in the reference buffer 2 is noise.
If it is noise, the characteristics of the characteristic pixel are referred to as a reference buffer 2.
It is preferable to add a noise elimination means 3 for elimination. In this case, an algorithm for determining whether or not the characteristic pixel TD is noise may be appropriately selected.

When extracting image features for a color image, it is preferable that the image signal combining means 4 combine the respective color component signals at an appropriate ratio and extract the characteristic pixel TD based on the combined signal. . In this case, it is preferable to appropriately select the combination ratio of each color component signal for each property of the characteristic pixel TD to be extracted so as to optimize the extraction accuracy of the characteristic pixel TD. Further, from the viewpoint of more accurately extracting the image features from the color image, the image signal input to the image signal synthesizing unit 4 is a uniform color space in which the color difference matches the human sense. It is preferable to map the input image and express each color component signal (brightness signal, chromaticity signal, saturation signal) in a uniform color space in a polar coordinate expression.

[0014]

According to the above-described technical means, as shown in FIG. 1, the characteristic pixel extracting means 1 converts a pixel of each scanning line into a highly versatile characteristic pixel TD, for example, an edge having a large density change. An edge feature pixel TDE corresponding to a pixel portion and having a rising edge and a falling edge, or a density gradient feature pixel TDN representing a density gradient of a pixel portion having a density gradient, or an edge feature pixel TDE and a density gradient feature pixel A feature pixel TD including TDN is extracted. Next, the characteristics of the characteristic pixel TD extracted by the characteristic pixel extracting means 1, such as the rising state or the falling state and the changing direction of the edge characteristic pixel TDE, and the density gradient direction of the density gradient characteristic pixel TDN are referred to. It is written for each corresponding pixel of the buffer 2 for use. As a result, the individual image processing unit refers to the property necessary for the image processing of the characteristic pixel TD for each pixel written in the reference buffer 2 and performs predetermined image processing on the pixel corresponding to the characteristic pixel TD. Can be applied. In particular,
When it is determined that the characteristic pixel TD written in the reference buffer 2 is noise, the noise elimination means 3 removes the noise, so that the credibility of the characteristic of the characteristic pixel TD written in the reference buffer 2 is reduced. Is very high, and accordingly, the image processing by the individual image processing means is accurately performed on the pixels that need to be processed.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the embodiments shown in the accompanying drawings. Embodiment 1 FIG. 2 shows an embodiment of a color image forming apparatus to which the present invention is applied. In FIG. 1, an input image read by a scanner or the like and subjected to appropriate preprocessing such as background removal is converted into polar coordinate display in a uniform color space as image data in consideration of a human visual system in an image input conversion unit 10. Is done.

Image input converter 1 used in this embodiment
0 is an input image buffer 11 for temporarily holding an input image
And an image prediction unit 12 that fetches an input image for each block from the input image buffer 11 and predicts in advance how the input image will look in a human visual system.
A data shaping unit 13 for appropriately converting input image data based on the prediction result of the image prediction unit 12;

FIG. 3 shows a specific processing example of the image prediction unit 12 and the data shaping unit 13. In the figure, the image prediction unit 12 is composed of a combination of, for example, three prediction units a1 to a3. In this embodiment, the prediction unit a1 predicts the distance (hereinafter referred to as an observation distance) because the appearance changes depending on how much the human looks away when viewing the document.
Since the color looks different because of the influence of other colors around the color (hereinafter, the color is referred to as a psychological color), the prediction amount is predicted by the prediction unit a3.・ Considering the O'Brien corn sweet effect,
This is to predict what humans can see. The data shaping unit 13 is composed of, for example, a combination of four shaping units b1 to b4, and performs necessary data shaping based on the prediction of the image prediction unit 12. In this embodiment, the shaping unit b1 applies a Gaussian filter to the input image based on the observation distance, and the shaping unit b2 converts the color of the input image into a psychological color. b3
Adds other illusion effects to the input image,
Furthermore, the shaping unit b4 is a set of ▽ ² G filters of appropriate size calculated from the observation distance (D. Marr, and E. Hildreth, “
Theory of edge detection, "Proc.R.Soc.Lond.B204, pp.
301-328) and apply them to the input image to create line drawings with several types of resolution. Then, the data shaping unit 13 performs necessary data shaping (b1 to b
3) Data shaping image DT1 (mapped to uniform color space in polar coordinates (L ^* a ^* b ^* )), line image DT2 having several resolving powers shaped by shaping section b4 and raw image Is output.

Next, the data shaped image DT1 of the image signals processed by the image input conversion unit 10 is always sent to the image processing unit 20, and a line image DT2 having several kinds of resolution and a raw input The image DT3 is sent to the image processing unit 20 as needed. In this embodiment, only the data shaping image DT1 is used in the image processing unit 2 to simplify the following description.
Assume to be sent to 0. In this embodiment, reference numeral 21 denotes image data from the image input conversion unit 10 (in this embodiment, a binary image is converted by means such as screen removal, etc.
It shall be converted to quantized continuous tone data.)
Is an image buffer that temporarily holds N. For example, in the case of processing N (positive integer) lines at a time, at least N + 2
It is necessary to have a storage capacity capable of holding multi-valued image data for a row.

Reference numeral 22 denotes a feature extracting unit for extracting features of an image based on image data, and reference numeral 23 denotes a feature extracting unit 22.
Are feature buffers for writing the feature data of the image extracted by... The detailed configuration of these is shown in FIG. In the figure, a feature extraction unit 22 converts each of the input image signal components into brightness, chromaticity, and saturation, and extracts an image feature from a signal obtained by linear synthesis of the components. A signal combining unit 221 is provided. The image signal synthesizing unit 221 appropriately switches coefficients for linearly synthesizing image signal components for each type of characteristic pixel to be selected. When extracting edge characteristic pixels, which will be described later, brightness (k1), The coefficients of the chromaticity (k2) and the saturation (k3) are α1, β1, and γ1, and the brightness (k1), the chromaticity (k2), the saturation (k)
The coefficients of 3) are α2, β2, and γ2. In the case of the present embodiment, as shown in FIG. 4, from the viewpoint of emphasizing the degree of influence of lightness and reducing the change in hue, α1,
α2 = 0.6, β1, β2 = 0, γ1, γ2 = 0.4.

Reference numeral 222 denotes a feature pixel extraction unit for extracting a feature pixel. In this embodiment, an edge feature pixel extraction unit 223 for extracting an edge feature pixel corresponding to an edge pixel portion having a large density change, and a density gradient And a density gradient characteristic pixel extraction unit 224 that extracts a density gradient characteristic pixel composed of a certain pixel portion.

The characteristics of the characteristic pixels extracted by the characteristic pixel extracting unit 222 are written in the characteristic buffer 23. The characteristic buffer 23 stores the edge characteristic pixel extracting unit 223 and the density gradient as shown in FIG. Corresponding to the characteristic pixel extraction unit 224, the memory device includes N + 2 row line buffers 231 and 232 each having a storage capacity of N + 2 rows. Also,
Reference numeral 225 denotes a noise deleting unit that determines whether the characteristic pixel written in the characteristic buffer 23 is noise, and deletes the noise if it is noise.

The individual image processing unit 24 (specifically, 2
4 (1) to 24 (n)) determine whether or not their own processing is necessary based on the information in the feature buffer 23, and if necessary, how strong the processing is. It decides and processes the image of the image buffer 21 according to it. This processing is sequentially repeated from the individual image processing unit 24 (1) to the individual image processing unit 24 (n), and finally the image buffer 21 contains an image on which necessary processing has been appropriately performed.
In this embodiment, the individual image processing unit 24 includes an edge enhancement process, an image cutout process, a color compression process, a noise removal process, and the like.

Thereafter, the processing is transferred to the image output unit 30,
Commonly known color correction 31 and density correction 3
After each process such as binarization and binarization 33 is performed, an image is output. Here, in evaluating the quality of the output color image, a type without the image prediction unit 12 and the data shaping unit 13 is used as a comparative example, and the image output in the embodiment and the image output in the comparative example are compared. In comparison, it was confirmed that the image reproducibility of the example was superior with respect to the outline of the image, the degree of shading, and the like.

Next, the process of extracting image features employed in the color image forming apparatus according to this embodiment will be described with reference to the flowchart of FIG. In the figure, first, the pointer of the characteristic buffer 23 is moved to a predetermined position (step 1), and the characteristic
Characteristic pixels are selected while scanning the required number of rows N of images input from 1 and, if there is a characteristic pixel, the type symbol of the characteristic pixel is recorded at a corresponding position in the characteristic buffer 23 (step 2). Next, it is determined whether or not the characteristic pixel selected by the noise removing unit 225 is noise based on whether or not there is another adjacent characteristic pixel. If not, the type symbol of the characteristic pixel is deleted from the position of the corresponding characteristic buffer 23 (step 3). After that, the processing of the input image is shifted to the N-th destination (step 4), and then the characteristic pixels from which noise has been removed are used in the individual image processing unit 24, or the characteristic pixels are transferred from the characteristic buffer 23 to another storage area. Is moved (step 5). Thereafter, it is determined whether or not all the rows of the input image have been scanned (step 6), and the processing of steps 1 to 5 is repeated until the processing of all the rows is completed.

Here, each processing will be described more specifically. Now, the case where an input image is stored at 400 spi and the number of gradations is 256, that is, one pixel is stored at 8 bits, and the image is processed two rows at a time (N = 2) will be described. The capacity of the buffer 21 and the feature buffer 23 requires at least N + 2 rows, that is, four rows. At this time, the extracted characteristic pixels have already been written in the uppermost row of the characteristic buffer 23, and noise removal has also been completed. Also, 2
In the row, the extracted characteristic pixels are written, but the noise has not been removed yet. Further, the third and fourth lines contain characteristic pixels that have already been extracted and used by the individual image processing unit 24, or have been copied to another storage area. Therefore, the newly extracted characteristic pixels are written in the third and fourth rows.

Next, a method of selecting characteristic pixels by the characteristic pixel extracting unit 222 will be described. Now, as shown in FIG. 7, it is assumed that the pixel of interest a is located at the center of the 3 × 3 pixel matrix, and the surrounding pixels are b, c, d, e, f,
As g, h, and i, calculations are performed according to the following equations (1) and (2).

[0027]

(Equation 1)

In the equation (2), Sum represents the difference between the pixel a and the average of the pixels b to i around the pixel.
In other words, if the value of Sum is small, there is no difference from the surrounding pixels, so that it cannot be a characteristic pixel. A specific example is shown in FIG.
(B). FIG. 8A shows a substantially uniform concentration distribution. In this case, Sum = -4.375, which is small as expected. FIG. 8B shows an example having a density gradient rising to the right. In this case, Sum = 1.75 also shows a small value. Although a general edge filter detects an edge, here it is necessary to perform finer processing. Therefore, a rising edge and a falling edge (beginning of falling) are distinguished, and a gradual density gradient is defined as an edge characteristic pixel. Will not judge. However, if it is desired to use the density gradient information in the subsequent individual image processing, the following equation 2 may be calculated.

[0029]

(Equation 2)

Even if Sum is small and it is not determined that the pixel is a characteristic pixel, if Abs is large in Equation 2, there is a density gradient.

The following equation (1) is adopted as a condition for judging an edge feature pixel. Also, even if Sum> CC1
May be noise even if The determination of noise is also performed by the noise removing unit 225 at the next stage, but is also performed by a simple method here. If Sum is extremely large, only the pixel of interest has low or high density, and it is highly likely that the noise is noise. Therefore, when the condition of Expression 3 (2) is satisfied, it is determined that the noise is present. Therefore, Sum
If CC1 <Sum <CC2, pixel a is determined to be an edge feature pixel. Further, when it is desired to detect the density gradient characteristic pixel, it is determined that the density gradient exists when the target pixel satisfies the following equation (3).

[0032]

(Equation 3)

Next, it is determined in which direction the edge feature pixel and the density gradient feature pixel have a feature, that is, which direction of the pixel a has a large density change. for that reason,
Max = max ｛| B |, | C |, | D |, | E |, |
F |, | G |, | H |, | I |} are determined to determine the direction of the density change, and the sign of the difference between the average value of the pixels in that direction and the pixel of interest a is used to determine the sign of the gradient. For example, FIG.
In the case of (c), the direction is e +. This is expressed as ↓ + here for convenience, and is recorded in the feature buffer 23. In the case of a density gradient characteristic pixel, since the directions → + and ← − mean the same, only the + direction is adopted, and an underline is added to indicate that this represents a density gradient → Is displayed.

FIG. 9 shows symbols corresponding to changes in image density. In the figure, two characteristic pixels → +, ← −
It can be seen that there is a large change in density between the two pixels where. Also, if the slope is gentle, it can be seen that some pixels have a density gradient symbol continuously, and in the case of a relatively steep change in density, the characteristic pixel is located at both ends or at either end. +, ←-should appear. Actually, it is necessary to determine CC1, CC1 ', and CC2 so that the image and the mark of the edge feature pixel / density gradient feature pixel correspond to each other in this manner.

Next, details of the noise removing unit 225 will be described. Now, in FIG. 6, since only the first line of the feature buffer 23 has been subjected to the noise determination, the data of the first line is used when processing the second and third lines. Here, there is no problem when noise determination is performed after the feature extraction for two rows is completed. However, when feature extraction and noise determination are performed for each pixel, the pixels are processed in the order shown in FIG. There is a need to continue to. That is, since the feature extraction of the surrounding pixels needs to be completed for the noise determination, the feature extraction needs to be completed for the pixel D5 in FIG. This means that the feature extraction must be completed up to the pixel D6.

If the target pixel is extracted as an edge feature pixel, it is checked whether there is a feature pixel having a directional component similar to the surrounding pixels as shown in FIG. If not, it is determined as noise and the edge feature pixel mark is deleted. Here, the similar direction components are directions in which the signs of the edge feature pixels are the same and are within ± 80 ° from the direction of the target pixel. For example, if the target pixel is ↓-, ↓-, ↓ (tilt 45 ° right)-, ↓ (tilt 45 ° left)-. Therefore, the target pixel in FIG. 10 is not a noise but an edge feature pixel.

Also, as shown in FIG. 12, even if the extracted edge feature pixel is isolated, there is a pixel that shows a density gradient in the direction in which the edge feature pixel is facing, and the edge feature pixel is +. If the concentration gradient in a similar direction is
In this case, a method of determining that the target pixel is truly a characteristic pixel even when a density gradient in the opposite direction is represented is adopted.

When the noise determination on the second and third lines of the characteristic buffer 23 is completed, the characteristic pixels written on the first and second lines of the characteristic buffer 23 become available. Is used by the individual image processing unit 24 or copied to another storage area. Then, the second row, that is, the first row and the second row are in a state where a new feature pixel mark can be stored again. Then, the processing result is written in these two lines. As described above, the feature extraction process is repeated up to the last line of the input image.

As described above, noise pixels are removed,
After organizing the information of the characteristic pixels, the individual image processing unit 24 performs desired image processing on the input image using the information of the characteristic pixels. Note that, depending on the contents of the individual image processing unit 24, it is necessary to configure the characteristic pixel extracting unit 222 that satisfies each requirement. Then, the image subjected to each individual image processing is output from the image output unit 30,
When this output image quality was compared with a comparative example (one in which an image was photographed, halftone dot, or text was used and individual image processing was performed), the image quality of the example was extremely excellent. It was confirmed that. Also, since the features of the image can be expressed in detail and the features can be used in common by various individual image processing units, the overall image processing unit configuration is simple, compact, and highly functional. can get.

Second Embodiment FIG. 13 shows a method of extracting edge feature pixels and density gradient feature pixels with higher accuracy than in the first embodiment. First, the first embodiment
Similarly, B, C... I
(See FIG. 13A), Sum and Abs are calculated according to the above equations 1 and 2, respectively, and pixels satisfying CC1 <Sum <CC2 are obtained. Next, the absolute values are compared, and the maximum value and the second largest value are extracted. Now, FIG.
As shown in (b), if the signs of the maximum value and the second largest value are the same and are adjacent to each other (80, 90 in the figure),
The target pixel a is determined to be an edge feature pixel. Next, in the case of extracting the density gradient feature pixels, Sum ≦ CC1 and Abs
Search for a pixel satisfying> CC1 '. Then, as shown in FIG. 13C, the sign of the maximum value and the sign of the second largest value are different from each other, and the pixels are adjacent to each other in a direction in which the target pixel is rotated by 180 degrees (99, -95 in the figure). Then, the target pixel a is determined to be a density gradient characteristic pixel.

[0041]

As has been described in the foregoing, according to the first aspect of the present invention, after extracting the one we feature pixel grayscale image data, writes the feature pixel in the reference buffer,
Since the image processing unit can be shared by a plurality of individual image processing units, the entire system of the image processing unit can be made compact. In particular, according to the present invention, the input image
Multi-gradation with correction processing based on human visual characteristics
Extract image features from image data
So how does the image appear in the human visual system?
Image features can be extracted while taking into account
The more accurate image features can be extracted,
Improved individual image processing accuracy and optimization of input image reproducibility
Can be achieved. Further, according to the invention of claims 2 4, wherein, in addition to being able to put together the entire system of the image processing unit in a compact, of multi-tone image data or we feature pixels which have been quantized, for example, rising And edge feature pixels that are distinguished from falling, or
Density gradient feature pixels, or feature pixels including edge feature pixels and density gradient pixels are extracted, and the characteristics of each feature pixel are written to the reference buffer. By referencing only the characteristics of the characteristic pixels written in the reference buffer to those required for individual image processing, for example, different processing can be performed according to the rising and falling edges of the image, and the density gradient of the image can be adjusted. Appropriate individual image processing at appropriate locations, such as performing different processing depending on the degree of image processing, or accurately grasping the volume of the image based on the edge portion of the image and the density gradient of the image and performing fine processing. can be applied by, Ru can improve the individual image processing accuracy.

According to the fifth aspect of the present invention, since the credibility of the characteristics of the characteristic pixels in the reference buffer can be further improved, it is possible to realize more accurate image characteristic extraction. Further, according to the invention described in claim 6 ,
In extracting the characteristic pixel, the surrounding pixel information is considered in addition to the information of the target pixel alone, so that the extraction accuracy of the characteristic pixel can be improved. Furthermore, claim 7
According to the invention described in any one of (9) to (9) , when performing image feature extraction on a color image, an optimal image signal can be synthesized for each property of the feature pixel to be extracted.
Extraction of characteristic pixels with high accuracy from a color image can be realized.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing a configuration of an image feature extraction device according to the present invention.

FIG. 2 is an explanatory diagram illustrating a first embodiment of a color image forming apparatus to which the present invention is applied;

FIG. 3 is a block diagram illustrating a specific example of an image input conversion unit according to the first embodiment.

FIG. 4 is an explanatory diagram illustrating details of an image processing unit according to the first embodiment.

FIG. 5 is a flowchart illustrating a specific processing example of an image processing unit according to the first embodiment.

FIG. 6 is an explanatory diagram illustrating an example of use of a feature buffer according to the first embodiment;

FIG. 7 is an explanatory diagram illustrating a relationship between a target pixel and surrounding pixels according to the first embodiment.

FIGS. 8A to 8C are explanatory diagrams illustrating examples of characteristic pixels extracted in the first embodiment.

FIG. 9 is an explanatory diagram illustrating an example of a correspondence relationship between shading of an image and characteristic pixels according to the first embodiment.

FIG. 10 is an explanatory diagram illustrating a processing order of pixels by a noise removing unit according to the first embodiment.

FIG. 11 is an explanatory diagram illustrating one mode in which noise is not deleted by the noise deletion unit according to the first embodiment.

FIG. 12 is an explanatory diagram showing another mode in which the noise is not deleted by the noise deletion unit in the first embodiment.

FIG. 13 is an explanatory diagram illustrating the principle of image feature extraction of the image forming apparatus according to the second embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Feature pixel extraction means, 2 ... Reference buffer, 3 ... Noise deletion means, 4 ... Image signal synthesis means, 5 ... Image prediction means, 6 ... Data shaping means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04N 1/40 G06T 1/00 510 G06T 7/60 250

Claims (9)

(57) [Claims] When extracting image features from quantized multi-tone image data , it predicts how an input image appears in human visual characteristics.
Then, based on this prediction result, human visual characteristics
Feature data through pre-processing of data processing based on
Multi-gradation image data for extracting an element is created, and then a plurality of individual image processing means (7)
Common After extracting the feature pixel (TD) available in the corresponding the extracted feature pixels plurality of individual image processing unit properties (TD) (7) reference buffer can be shared (2) An image feature extraction method, wherein each of the image features is written at a corresponding pixel position. 2. An image feature extracting apparatus for extracting an image feature from quantized multi-tone image data, wherein the image feature extracting device is a pre-processing means for the multi-tone image data for extracting the image feature.
To predict how the input image will appear in human visual characteristics.
Image predicting means (5) for measuring, and the image predicting means (5)
Input image based on the human visual characteristics
Data shaping means (6) for performing data processing according to
So, from the pixels of each scan line, wherein the pixel extracting means for extracting a feature pixel at least corresponding to the large edge pixel portion of density change and the rise and fall and distinguish the edge feature pixels (TDE) is included (TD) (1) The property of the characteristic pixel (TD) extracted by the characteristic pixel extracting means (1) is written for each corresponding pixel and provided so that a plurality of individual image processing means (7) can share it. And a reference buffer (2). 3. An image feature extraction device for extracting image features from quantized multi-tone image data, wherein said image feature extraction device is a pre-processing means for multi-tone image data for extracting image features.
To predict how the input image will appear in human visual characteristics.
Image predicting means (5) for measuring, and the image predicting means (5)
Input image based on the human visual characteristics
Data shaping means (6) for performing data processing according to
So, from the pixels of each scan line, at least a concentration gradient characteristic pixels representing the gradient of the pixel portion of the gradient (TDN) feature pixel extracting means for extracting a feature pixel (TD) that contain (1), this feature A reference buffer (2) provided so that the property of the characteristic pixel (TD) extracted by the pixel extraction means (1) is written for each corresponding pixel and shared by a plurality of individual image processing means (7). An image feature extraction device comprising: 4. An image feature extracting apparatus for extracting an image feature from quantized multi-tone image data, wherein the image feature extracting device is a pre-processing means for the multi-tone image data for extracting the image feature.
To predict how the input image will appear in human visual characteristics.
Image predicting means (5) for measuring, and the image predicting means (5)
Input image based on the human visual characteristics
Data shaping means (6) for performing data processing according to
From the pixels of each scanning line, feature pixels including at least an edge feature pixel (TDE) corresponding to an edge pixel portion having a large change in density and a density gradient feature pixel (TDN) representing a density gradient of a pixel portion having a density gradient. A characteristic pixel extracting means (1) for extracting (TD); a plurality of individual image processing means in which the characteristic of the characteristic pixel (TD) extracted by the characteristic pixel extracting means (1) is written for each corresponding pixel; (7) An image feature extraction device comprising: a reference buffer (2) provided so as to be shared. 5. A as described in any one of claims 2 to 4, characterized pixels written in the reference buffer (2) (TD) is determined whether or not the noise, the feature pixel if noise An image feature extraction device characterized by adding a noise removing means (3) for removing the property of the reference from the reference buffer (2). 6. A as described in any one claims 2 to 4, characterized pixel extraction means (1) refers to the plurality of pixels located around the target pixel, the characteristic pixel of the target pixel (TD) An image feature extraction device characterized by extracting. 7. An image from quantized multi-tone image data
An image feature extraction device for extracting features, wherein each scanning line
Edge pixels with at least large density changes
Corresponding to the part and distinguished between rising and falling
Feature pixel (TD) including edge feature pixel (TDE)
Pixel extracting means (1 ) for extracting
The characteristic of the characteristic pixel (TD) extracted by the output means (1) is
A plurality of individual image processing steps are written for each corresponding pixel.
Reference buffer provided so that stage (7) can be shared
(2) an image corresponding to a color image
The feature extracting apparatus further includes image signal combining means (4) for combining respective color component signals in a uniform color space of each pixel at an appropriate ratio for each property of a feature pixel (TD) to be extracted. Feature extraction device. 8. An image from quantized multi-tone image data
An image feature extraction device for extracting features, wherein each scanning line
From the pixels with the density gradient at least
Including a density gradient feature pixel (TDN)
Characteristic pixel extracting means (1) for extracting a feature pixel (TD);
The characteristic pixel extracted by the characteristic pixel extracting means (1)
(TD) property is written for each corresponding pixel and multiple
Are provided so that the individual image processing means (7) can be shared.
And a reference buffer (2)
An image feature extracting apparatus for an image includes image signal combining means (4) for combining respective color component signals in a uniform color space of each pixel at an appropriate ratio for each property of a feature pixel (TD) to be extracted. Image feature extraction device. 9. An image from quantized multi-tone image data
An image feature extraction device for extracting features, wherein each scanning line
Edge pixels with at least large density changes
Edge feature pixel (TDE) and density gradient corresponding to the part
Density characteristic pixel (T
DN), and the characteristic pixel for extracting the characteristic pixel (TD)
Extraction means (1) and the characteristic pixel extraction means (1)
The property of the output characteristic pixel (TD) is written for each corresponding pixel.
And can be shared by a plurality of individual image processing means (7).
And a reference buffer (2) provided to
Of image feature extraction devices for color images
And an image signal synthesizing means (4) for synthesizing each color component signal in a uniform color space of each pixel at an appropriate ratio for each property of the characteristic pixel (TD) to be extracted. .