JP3909604B2 - Image processing apparatus and image processing method - Google Patents

Description

  The present invention relates to an image processing apparatus, and more particularly to an image processing apparatus that performs binarization and edge detection of an image.

  In the full-fledged information age, images are frequently handled by familiar electronic devices. In addition, the spread of computers and application software has made it easy for anyone to process digital images.

  When performing image processing, edge detection and binarization processing are often performed (see Patent Document 1 below).

  Binarization is an operation of converting the brightness of each pixel constituting an image into two values of black and white according to a certain reference value. Here, a certain reference value is referred to as a threshold value. Usually, each pixel of the image has an RGB value of 0-255. The average value of the RGB values is the brightness at each pixel. The binarized image differs depending on the reference threshold value.

Edge detection is an operation for calculating the contour of an object based on a change in the brightness of an image. All the pixels in the image have color information. The place where the color information in the image changes rapidly is the boundary (outline) of the object area. The change in brightness can be calculated by using differential calculation.
Japanese Patent Laid-Open No. 05-300369

  However, in a general binarization processing method, for example, in the case of an image with intense shading as shown in FIG. .

  Specifically, in order to obtain a binarized image from a grayscale image, it is necessary to set a boundary (threshold value) for determining 1 and 0. As a threshold setting method, there are a fixed threshold method in which all processes are binarized with one threshold and a floating threshold method in which the threshold is partially changed. When an image with severe shading as described above is processed by the fixed threshold method, an image as shown in FIG. 12B is obtained, and some information is lost. Here, the part where the information by the binarization process is lost is shown by hatching. Therefore, a floating threshold method must be used to select a partially optimal threshold. However, as shown in the histogram of FIG. 13, it is difficult to search for the bimodality that determines the threshold value in the density distribution of the image shown in FIG.

  Further, when general edge processing using differential processing is performed, the differential signal itself is small in a dark portion of an image, so that there is a problem that it is difficult to perform edge processing in this portion.

  The present invention has been made in view of the above problems. A main object of the present invention is to provide an image processing apparatus capable of performing edge processing and binarization processing satisfactorily even for images accompanied by shading.

  The image processing apparatus of the present invention is an image processing apparatus that performs image processing of the image using a coefficient based on a luminance value of a pixel included in the processing target image, and each of the images is a target pixel to be calculated. And dividing means for dividing the block into a plurality of blocks, and calculating an average luminance value by averaging the luminance values of the pixels included in the block in each of the blocks, and calculating the average luminance value of the target pixel and the average of the blocks Normalizing means for calculating the coefficient normalized using the luminance value.

  The image processing method of the present invention is an image processing method for performing image processing of the image using a coefficient based on a luminance value of a pixel included in the image to be processed. A division step of dividing the block into a plurality of blocks, and calculating an average luminance value by averaging the luminance values of the pixels included in the block in each of the blocks, and calculating the average luminance value of the target pixel and the average of the blocks And a normalization step of calculating the normalized coefficient by comparing with a luminance value.

  According to the present invention, even when binarization or edge detection of an image accompanied by intense shading is performed, these processes can be performed by almost completely removing low-frequency components caused by shading. Therefore, it is possible to perform binarization or edge detection of a document with low image quality.

<First Embodiment>
In this embodiment, the unevenness coefficient used when image processing is performed using the image processing of this embodiment will be described.

  The unevenness coefficient is a function obtained by normalizing density unevenness information of an image to be processed. By using this coefficient, image processing of an image whose brightness changes locally can be appropriately performed.

  Specifically, the unevenness coefficient is a value calculated by a pixel block unit divided into several pixels square unit, and each pixel value divided by an average luminance value of all pixels in the block. Therefore, this coefficient is obtained by normalizing the distribution state of the AC component in the block with the average luminance value of the block, and is information in which the DC component and the low frequency component are excluded. As a result, even if the intensity of illumination changes and the average brightness of the screen changes, the unevenness coefficient of the same portion on the screen can always be kept at the same value. In addition, even when the brightness on the screen changes gradually depending on the location due to factors such as illumination unevenness, the change in brightness is small enough to be ignored in a small area range such as a pixel block.

  Therefore, the unevenness coefficient can almost completely remove a low frequency component caused by illumination unevenness. This is different from the conventional method in which edge detection is performed by differential processing. In the conventional differential processing, even if the screen has the same uneven state, the differential signal itself becomes small and difficult to detect when the illumination becomes dark. For this reason, edge detection and binarization processing often do not work well with dark images or images with shading.

  Next, a method for calculating the unevenness coefficient will be described. The basic concept of this unevenness coefficient is to divide an image into n × n pixel units and obtain the ratio of the average value of all the pixels in the block and each pixel value by using the following equation 1 for each block. Value. Therefore, the unevenness coefficient is calculated for all pixels.

Here, X _ij is the j-th pixel value of the i-th pixel block composed of n × n pixels, Ave _i is the average luminance value of all the pixels in the i block, and r _ij is the unevenness corresponding to X _ij It is a coefficient.

  FIG. 1 shows the frequency distribution of the unevenness coefficient derived for a certain image. It can be seen that the unevenness coefficient values are concentrated in the vicinity of 1.0, and express the characteristics of the image such as the area of the flat portion of the flat portion of the image and the state of the edge portion.

  However, since the unevenness coefficient obtained by the method of Equation 1 normalizes each pixel value with the average value of the pixel blocks, block distortion may be detected when the block size increases. Therefore, for the purpose of suppressing block distortion that causes image quality degradation, a method for calculating the unevenness coefficient using the correlation with adjacent blocks was examined.

  A method for calculating a concavo-convex coefficient (hereinafter referred to as a smoothed concavo-convex coefficient) using a correlation with an adjacent block will be described.

  FIGS. 2 and 2 show a specific calculation method of the smoothing unevenness coefficient when the block size is 2 × 2.

Incidentally, X ₂₂ in this case is the calculation target pixels of the smoothing unevenness factor. As shown in Equation _2, pixel by pixel in the vertical and horizontal directions so as to include the _{X 22,} the four blocks of 2 × 2 pixels cut out while shifting, respectively to calculate the average value Ave1~Ave4, Ave1~Ave4 in the four ratios determined by dividing the X _22, to calculate the smoothing unevenness coefficient R ₂₂ by further averaging.

  However, in this method, as the divided block size is increased, the number of divisions for calculating the average value and the smoothing unevenness coefficient increases and the calculation time becomes longer. Therefore, a simple calculation method for calculating the average value approximately was examined. This takes into account the overlapping number of pixels used for the calculation of the four average values, and Equation 3 shows the calculation formula. In this method, since the average value can be calculated by one calculation, the unevenness coefficient can be derived easily and quickly. In addition, this calculation formula can also suppress duplication of rounding errors due to division.

The average value calculation method of Equation 3 is a calculation similar to convolution integration using a 3 × 3 mask operator. That is, the smoothing unevenness coefficient is a value obtained by dividing the target pixel value as shown in FIG. 3 by the weighted average value. Therefore, the unevenness coefficient derived by this method was defined as a smoothed unevenness coefficient, and a simulation experiment was performed using this calculation method thereafter.

  If the divided block size is increased to 3 × 3 or 4 × 4, the calculation range for deriving the unevenness coefficient is expanded to 5 × 5 and 7 × 7 pixel sizes, respectively, and the calculation range is wide. Become. As an example, FIG. 4 shows a mask operator when the pixel block size is 3 × 3.

  The characteristics of the smoothing uneven function will be described. As described above, the smoothing unevenness coefficient is obtained for each pixel block divided into units of several pixels square, and is a value obtained by calculating the ratio between the average luminance value of all the pixels in the block and each pixel value. Therefore, this coefficient becomes a function representing only the density change rate in the micro portion of the image, and the DC component of the pixel block is automatically excluded. As a result, the unevenness coefficient of the same pixel portion on the screen always has the same value regardless of the illumination state.

  In addition, when the change in brightness on the screen is a small change that can be ignored when considered in units of pixel blocks, the change in brightness can be almost completely eliminated. This is one of the great advantages of the smoothing unevenness coefficient. If this feature is used, non-uniform low frequency information such as illumination unevenness can be removed during edge detection and binarization processing. Moreover, since the convolution integral which considered the correlation between adjacent pixels is used for the calculation of the smoothing uneven | corrugated coefficient, the discontinuity between blocks can be suppressed.

<Second Embodiment>
In this embodiment, an example in which image processing is specifically performed using the smoothing unevenness coefficient described in the first embodiment will be described.

  First, an example in which edge components are extracted using a smoothing unevenness coefficient will be described.

  In order to confirm the characteristics of the smoothing unevenness coefficient, an image quantized with 8 bits shown in FIG. 5 was selected as the target image, the smoothing unevenness coefficient was calculated, and extraction of edge components was attempted. The mask operator was 3 × 3. An edge image was produced by multiplying the calculated smoothing unevenness coefficient by a value of 128, which is an intermediate value of 8 bits (256 levels), as a bias. The image is shown in FIG.

  FIG. 7 shows an edge image processed by an 8-direction Laplacian filter (bias value 128) having a 3 × 3 pixel size as a comparison target image. When these two images are compared, the edge detection effect does not change, but the image processed with the smoothing unevenness coefficient has less ringing and fine graininess noise present in the processed image of the Laplacian filter, and the image is clear. I can see that This is presumably because the smoothing unevenness coefficient performs an average value calculation in consideration of the peripheral image, and thus has an integral effect on the screen.

  Next, edge extraction of an extremely dark image with illumination as shown in FIG. As in the previous case, the smoothed unevenness coefficient is calculated from the original image of FIG. 8A, and the edge image obtained by multiplying the coefficient by 128 is shown in FIG. 8B, and the image processed by the Laplacian filter is shown in FIG. Shown in (C).

  The image shown in FIG. 8B is subjected to edge processing relatively clearly. On the other hand, in the image shown in FIG. 8C, a clear edge is not extracted. That is, the edge detection method using the smoothing unevenness coefficient indicates that the edge detection capability does not deteriorate even in a dark screen. This is because the smoothing unevenness coefficient is not influenced by a gradual change in brightness of the screen, and the contrast ratio of the micro portion of the screen can be faithfully reproduced. Such a feature is difficult to realize by the conventional differential edge detection method, and can be said to be a large feature of the smoothing unevenness coefficient.

  Next, an example in which binarization processing is performed using a smoothing unevenness coefficient will be described.

  The smoothing unevenness coefficient always normalizes the average value of the pixel block, which is a micro part of the image, to 1. Therefore, in the case of the smoothing unevenness coefficient, the threshold level of the pixel block can be automatically changed by setting only one threshold. That is, an optimum threshold value is automatically set for each pixel block, and binarization processing can be performed without adverse effects such as uneven illumination.

That is, as shown in the following equation 4, the variation threshold value binarization process can be performed by limiting the range of the smoothing unevenness coefficient R _ij .

FIG. 9 shows the result of binarizing the shading image using this method. FIG. 9A shows an image before binarization processing, and FIG. 9B shows a processed image obtained by binarization processing. As shown in the processed image, the binarization processing by this method clearly detected characters over the entire surface, and a result that hardly felt the influence of shading was obtained. In this way, the binarization processing using the unevenness coefficient operates as a floating threshold method that automatically selects the optimum threshold value by setting only one threshold value for the coefficient value, and is excellent for any image. The binarization processing capability can be demonstrated.

  Next, with reference to FIG. 10, the character thickness and block size will be described in the case of processing an image including characters.

  In the binarization processing method using the concavo-convex coefficient, the direct current component of the image to be processed is deleted. Therefore, when the thickness of the character is larger than the pixel block size, white as shown in FIG. Omission problems occur. This occurs because the concavo-convex coefficient of a white screen and a black screen are both 1. That is, when the threshold is set to 1 and binarization is performed under the condition of Equation 4, when the entire pixel block enters a black region and the unevenness coefficient becomes 1, the black region is inverted to white.

  As a solution to this problem, there is a method of making the pixel block size larger than the thickness of the character. An image binarized by this method is shown in FIG. This image has no white spots.

As another method for the above problem, there is a method of monitoring the average luminance value of the pixel block. Specifically, the average luminance value Ave _i of the pixel block is constantly monitored, and even if the unevenness coefficient is 1, if the Ave _{i is} almost 0, that portion is forcibly made black. Even with this method, an image as shown in FIG. 10B can be obtained.

<Third Embodiment>
In this embodiment, two types of image processing filters using the above-described unevenness coefficient or smoothed unevenness coefficient will be described. The first filter is a filter whose frequency characteristics do not change with changes in brightness, and the second filter is a filter whose frequency characteristics change according to changes in brightness.

  First, the calculation formula of the first filter described above is shown in Equation 5 below.

In Equation 5, Ave _ij is the average value of the pixels of the block, and R _ij is the unevenness coefficient or the smoothed unevenness coefficient. Y _ij is a pixel value after filtering.

  In this filter, the filter characteristics can be freely changed by an arbitrary constant n. In addition, a constant frequency characteristic can always be obtained regardless of the change in brightness of the image. This linear filter can freely change the characteristics depending on the value of n from edge enhancement to blurring characteristics. The characteristic of the first filter changes with n = 1 as a boundary. Specifically, when n is larger than 1, the first filter performs edge enhancement processing. When n is smaller than 1, the first filter performs a blurring process. When the value of n is 1, it is expanded as shown in Equation 6 below, and the pixel value after filtering is the same as the input pixel value.

Next, a second filter whose frequency characteristics automatically change according to the brightness of the screen is shown in the following equation (7).

In Equation 7, Ave _ij is an average value of the pixels of the block, and X _ij is an input pixel value. In this filter, the degree of edge enhancement and blurring can be freely changed by an arbitrary constant k. In this filter, the frequency characteristic of the filter automatically changes depending on the magnitude relationship between the value of the parameter k and Ave _ij . Specifically, the frequency characteristic of the filter changes with k = Ave _ij as a boundary. When the value of k is larger than Ave _ij , it has edge enhancement (high pass characteristics) characteristics. Further, when the value of k is smaller than Ave _ij , it has a blurring process (low-pass characteristic) characteristic. Further, when k = Ave _ij, the following expression 8 is developed, and the pixel value after the filter processing becomes equal to the input value.

In this filter, the frequency characteristic automatically changes from a high-pass characteristic to a low-pass characteristic due to a change in image brightness (block average value).

  The image shown in FIG. 11A is an image in which the brightness is changed so that the left side is brighter than the right side.

  FIG. 11B is an image obtained by processing the image shown in FIG. 11A with the second filter. Referring to this image, it can be understood that the frequency characteristics change as the brightness changes. That is, edge-enhanced image processing is performed on the right side of a low-lightness (dark) image. In addition, blurring processing is performed on the left side of an image with high (bright) brightness.

  With this filter, edge enhancement is performed in a dark portion of an image. Therefore, for example, by emphasizing an edge of a dark part, a lesion part of the X-ray photograph can be emphasized.

It is a graph which shows the histogram of the uneven | corrugated coefficient of this invention. It is a table | surface which shows the calculation range of the smoothing uneven | corrugated coefficient of this invention. It is a table | surface which shows the mask operator (2x2) used when calculating the smoothing uneven | corrugated coefficient of this invention. It is a table | surface which shows the mask operator (3x3) used when calculating the smoothing uneven | corrugated coefficient of this invention. This is an original image to be subjected to edge processing. It is the image which performed edge detection using the image processing apparatus of this invention. It is the image which performed edge detection using the conventional differentiation method. (A) is an original image to be subjected to edge processing, (B) is an image subjected to edge detection using the smoothing unevenness coefficient of the present invention, and (C) is a conventional differentiation method. It is the image which performed edge detection. (A) is an image to be binarized, and (B) is an image that has been binarized using the smoothing unevenness coefficient of the present invention. (A) and (B) are images binarized using the smoothing unevenness coefficient of the present invention. (A) is an image before performing the filtering process, and (B) is an image after performing the filtering process of the present invention. (A) is an image to be binarized, and (B) is an image to which a conventional binarization process is applied. 13 is a graph showing a histogram of the image shown in FIG.

Claims (7)

  An image processing apparatus that performs image processing of the image using a coefficient based on a luminance value of a pixel included in an image to be processed;
  Dividing means for dividing the image into a plurality of blocks each including a target pixel to be calculated;
  In each of the blocks, the average luminance value is calculated by averaging the luminance values of the pixels included in the block, and normalized using the luminance value of the target pixel and the average luminance value of each block Normalization means for calculating the coefficients;
  An image processing apparatus comprising:   The normalizing means obtains a ratio by dividing the luminance value of the target pixel by the average luminance value of each of the blocks, and calculates the normalized coefficient by averaging the ratio of each of the blocks. The image processing apparatus according to claim 1, wherein:   The dividing means includes a first block including the target pixel at a lower right end, a second block including the target pixel at an upper right end, and a third block including the target pixel at an upper left end. , Dividing the target pixel into a fourth block including the lower left portion,
  2. The normalization unit calculates the coefficient from luminance values of pixels included in each of the first block, the second block, the third block, and the fourth block. Image processing apparatus.   The image processing apparatus according to claim 1, wherein binarization of the image, edge processing of the image, or filtering of the image is performed using the coefficient.   The image is an image including characters,
  The size of the block is made larger than the line constituting the character, and binarization using the coefficient is performed,
  The image processing apparatus according to claim 1, wherein characters are detected from the binarized image.   Binarization means for providing a black signal indicating black or a white signal indicating white based on the coefficient;
  The image processing apparatus according to claim 1, further comprising: a changing unit that changes a white signal provided by the binarizing unit to a black signal based on an average luminance value of the block.   An image processing method for performing image processing of the image using a coefficient based on a luminance value of a pixel included in an image to be processed;
  A division step of dividing the image into a plurality of blocks each including a target pixel to be calculated;
  In each of the blocks, the luminance value of the pixels included in the block is averaged to calculate an average luminance value, and the luminance value of the target pixel and the average luminance value of each block are compared and normalized. A normalization step of calculating the coefficient;
  An image processing method comprising: