KR100389893B1 - Apparatus for coding image by using image feature classification system - Google Patents

Abstract

PURPOSE: An apparatus for coding an image by using an image feature classification system is provided to classify the image complexity by using an original image, and classify and remove perceptual prediction errors according to the complexity, thereby improving the image coding efficiency. CONSTITUTION: A perceptual error processing unit(200) includes an image complexity classifier(300), a tolerance perceptual color adjusting unit(302), and a perceptual prediction generator(304). The image complexity classifier receives a pixel 0(x,y)(0<=x<=N-1, 0<=y<=N-1) of an original image for defining the image complexity of an MxM window by using color perception. The image complexity classifier grants a PCT(Perceptual Color Tolerance) weight to pixels to be coded. The tolerance perceptual color adjusting unit receives the complexity for dividing the same into complexity intervals for granting image features to an encoder by granting PCT weights according to the complexity. A perceptual prediction generator compares the PCT weights with the sizes of color components of a predictive error image, and substitutes 0 for the predictive errors smaller than the PCT weights, thereby defining a perceptual prediction error image.

Description

Image Coding Apparatus Using Image Feature Classification

The present invention relates to an image encoding apparatus using an image characteristic classification method, and to improve image coding efficiency by removing an error that cannot be visually sensed from a motion prediction error using an image complexity and an allowable gamma color difference, which are defined using a gaze color difference characteristic. An image encoding apparatus using an image characteristic classification method.

In the video encoding technique aimed at minimizing the amount of encoded image data and minimizing deterioration of the reconstructed image, the correlation between the image data between the previous image and the current image in the time domain and the correlation between the image data in the spatial domain A compression encoding method is used to reduce the amount of image data to be encoded while minimizing the degree of deterioration of the quality of the reconstructed image. In particular, in the time domain, a prediction error image is constructed by using a motion between an image at a previous time and an image at a current time, and a method of encoding and compensating the prediction error image is generally used. On the other hand, among the image prediction errors used to increase the correlation between the image data, there are actually errors that cannot be distinguished by the naked eye. If only such prediction errors are selected and removed, the image encoding efficiency can be further increased.

In the conventional digital image encoding method, two encoding techniques are generally used as shown in FIG. 1 as a method of increasing encoding efficiency by using correlation between input image data.

1 is a block diagram of a conventional video encoding apparatus.

The input image storage unit 100 stores the digital image (t) to be encoded currently and the original images in a before or after time slot used to estimate a motion vector in the motion estimation unit. The motion estimator 102 finds a block having an image characteristic most similar to that of the current image block n to be encoded using a previous original image or a subsequent original image, and transmits position information about the block, that is, a motion vector to the encoder. Function The image to be encoded is processed in units of 16 × 16 image blocks. The motion estimator 102 uses the difference value between the predicted video block, which is estimated to have the most similar video data property to the video block n to be encoded, as the video signal data to be encoded, thereby calculating the amount of video data to be encoded. It will have a reducing effect. The motion compensator 104 obtains the image block of the place designated by the motion vector value of the n-th original block to be encoded obtained by the motion estimation unit 102 from the previous reconstructed image stored in the image restorer 106. Do this. The image restorer 106 restores and stores the previous encoded images so that the motion compensator may define a predictive image block of the image block to be currently encoded using the motion vector. When the prediction error calculating unit 108 receives the nth encoded image O from the input image storage unit 100, the predicted error calculating unit 108 compares the previous image encoded by the n−1th encoding with the motion degree of each encoded image unit using a vector value. The prediction image P is generated for the original n-th image to be encoded using the estimated motion vector. The discrete cosine transforming unit 110 defines information as a characteristic value for each frequency component in a difference image between the reconstructed image block obtained by the motion compensation unit 104 by the n th original image block to be encoded and the motion vector. The quantizer 112 quantizes the DCT coefficients generated by the discrete cosine transform unit 110 to compress a real image. The variable length encoder 114 performs a function of constructing a bit stream used in the decoder according to the values of the DCT coefficients quantized by the quantizer 112. The buffer 116 performs a function of buffering the signal output from the variable length encoder 114. The inverse quantizer 118 and the inverse discrete cosine transforming unit 120 define an inverse quantization process, reconstruct the current encoded nth image block, and store it in the image restoring unit 106 to store the (t + 1) th circle. A function of providing a t-th reconstructed image for motion compensation when encoding an image is performed. The bit adjuster 122 adjusts the quantization size value according to the buffered amount of the buffer 116 in order to accurately match the actual total bit generation amount of the encoded image with the total target bit generation amount.

The first method is a transform encoding method that defines correlation of image data in a spatial frequency domain by converting input image data to be encoded using a discrete cosine transform (DCT) from a time frequency domain to a spatial frequency domain. This encoding method is characterized by extracting only spatial frequency characteristics of the image data without loss of the image data and using the same for the compression and encoding method of the image.

The second method is a predictive coding method using differential pulse code modulation (DPCM) to define the correlation of image data to be encoded in the time domain. In this method, as shown in FIG. 1, when the image input unit receives an n-th coded image (0), the motion degree of each encoded image unit is estimated by using a vector value, compared to the n-1th previous image. The predicted image of the original image to be n-th encoded is configured by the estimated motion vector. The predicted image configured as described above has an error due to movement when compared with the actual nth original image. The error caused by such movement is called a prediction error (e) and is defined as Equation (1).

e = O-P

The prediction error defined as in Equation 1 refers to the degree of correlation of the image data in the time domain. Since only the image data configured by the prediction error is used in the actual encoder, there is a loss of the image data, but the encoding efficiency of the image can be improved. Has the characteristics. Therefore, in the moving picture coding method, as shown in FIG. 1, a mixed coding method using both a discrete cosine transform (DCT) for extracting spatial correlation and a DPCM method for extracting temporal correlation is frequently used.

On the other hand, if human color perception is considered in the correlation extraction method of the image data, a higher efficiency encoding effect can be expected. For example, among the prediction error values obtained through the DPCM process, there are a considerable amount of prediction errors that cannot be identified by the human eye. Therefore, if the prediction errors are eliminated by using the PCT, the image quality of the same reconstructed image may be provided, but only the correlation of the image data may be relatively increased, thereby improving the coding efficiency. In addition, when the image to be encoded is a complex image, there is a visual perception characteristic in which the discriminating power of an error is relatively lower than that of a simple image. It is possible to further define the threshold value used to select an unidentifiable prediction error among the image errors predicted from the encoded images classified into the larger value as a larger value, thereby further increasing the effect of removing the prediction error.

An object of the present invention is to improve the image coding efficiency by classifying the image complexity using the original image by classifying the human visual characteristics as a look-up table for real-time processing, and classifying and removing visual prediction errors according to the classified complexity. The present invention provides an image encoding apparatus for rendering.

1 is a block diagram of a conventional video encoding apparatus.

2 is a block diagram of a video encoding apparatus according to the present invention.

3 is a detailed block diagram of the visual error processing unit 200 of FIG. 2.

4 is a diagram illustrating the visual perception color difference look-up table.

5 is an image complexity classification diagram.

6 is a diagram for comparing error image data and visual error image data.

The apparatus of the present invention for achieving the above object is configured to give a prediction error image e differently for each pixel to be coded PCT weight according to the image complexity to configure a gaze error image S consisting of only errors that generate a color gamut In the video encoding apparatus for improving the temporal correlation of the image data, the pixels O (x, y) (0≤x≤N-1, 0≤y≤N-1) of the original image are substituted instead of the variance using the Y component. M × M window centering on this pixel An image complexity classification unit for defining an image complexity of M-1 is a multiple of 4 by using a luminous color difference characteristic to give a PCT weight considering a luminous characteristic to a pixel to be encoded; The complexity defined by the image complexity classification unit is input and divided into complexity intervals of a specific size. If the complexity is large, a large PCT weight can be given otherwise or a small PCT weight can be differentially assigned according to the image complexity to encode an image to be encoded. Allowable time difference difference adjusting unit to provide a characteristic; And comparing the PCT weights (Ty, Tu, Tv) defined by the permissible color perception difference adjusting unit with the size (ey, eu, ev) for each color component of the prediction error image e, the prediction errors smaller than the PCT weight have no visual perception color difference. It includes a luminous error processing unit having a luminous prediction generating unit for defining a luminous prediction error image S according to luminous color difference characteristics by determining all errors as 0.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

2 is a block diagram of a video encoding apparatus according to the present invention.

In FIG. 2, components except for the visual error processing unit 200 are the same as those of FIG. 1, and thus description thereof is omitted.

The visual error processing unit 200 of FIG. 2 classifies image complexity using a look-up table modeling visual color difference characteristics of the prediction error image data before being input to the discrete cosine transform unit 110 and predicts the visual error that cannot be visually recognized. Remove them.

3 is a detailed block diagram of the visual error processing unit 200 of FIG. 2.

In Fig. 3, reference numeral 300 denotes an image complexity classification unit, 302 denotes a permissible visual color difference adjusting unit, and 304 denotes a visual perception prediction error generating unit.

First, the image complexity classification unit 300 will be described.

One pixel looks different depending on the surrounding image characteristics. That is, when the color of the surrounding image is distributed in various ways or the image is complicated, even if the value of this pixel is slightly changed, it is not easily noticeable. It is a processing part which modeled such a visual characteristic using the visual characteristic characteristic value. In this case, the color perception characteristic used refers to the maximum and minimum values for each of the Y, U, and V values that are not visually recognized when the Y, U, and V color values of one pixel are changed.

The permissible time difference difference adjusting unit 302 will be described.

The image complexity classifier 300 is a processor that adjusts color perception characteristic values of pixels according to characteristics of the complexity of the surrounding image for each pixel.

The visibility prediction error generation unit 304 will be described.

Among the error image data defined as the difference between the original image and the reconstructed image in the image encoding method, error signal data having an absolute value smaller than the color perception characteristic value defined by the permissible perceptual chrominance adjustment unit 302 may be identified by the naked eye. Since this error is eliminated, only the error data that is actually sensed visually is sent to the discrete cosine conversion unit 110 by removing the error. This reduces the amount of image data to be encoded without degrading the quality of the reconstructed image, thereby increasing the encoding efficiency.

The operating principle of the present invention is as follows.

The image complexity classifier 300 receives an N × N original image block O composed of Y, U, and V color values, and receives each image pixel O (x, y). , A lookup-table is used to determine the PCT value for and classify the surrounding image complexity of each pixel in the input image block into k types using the PCT. First, the look-up table for the PCT used in this processing unit applies each color component (L, a) by applying the CMC (Color Measurement Committee) visual color difference formula defined in Equation 2 on the CIE L * a * b uniform color gamut color coordinate system. , b) allowable time-sensitive color difference size is converted into a Y, U, V color coordinate system and configured as shown in FIG.

The method of defining the PCT values Ty, Tu, and Tv using the look-up table based on the color values Oy, Ou, and Ov of the input image pixel is expressed by Equations 3 and 4 below.

index = (Oy / 32) * 64 + (Ou / 32) * 8 + (Ov / 32)

Equation 4 defines Tu and Tv when the image input form is 4: 4: 4. The method of determining Tu and Tv is defined differently according to the input form of the image. For example, in the case of a 4: 2: 0 image input form in which one u and v components are combined with four Y components, four Tu values are defined for one u component by applying Equations 3 and 4 above. In this case, the minimum of four Tu values is determined.

In this way, when the PCT values for each pixel of the input image are defined, the complexity of surrounding images centered on each pixel in the input image block must be determined. As shown in (a) of FIG. For the N × N image, the determination is made using the color difference characteristics between adjacent pixels. That is, an image block to be encoded is defined using color difference comparison values for two directions.

M × M size windows The color difference magnitude vE with the horizontally adjacent pixels (i, j + 1) for one pixel W (i, j) and the color difference magnitude vE with the horizontally adjacent pixels (i-1, j) After the calculation, the pixels (i, j) are compared with the color difference size pE for the PCT values (Ty, Tu, Tv) of the pixels (i, j). The complexity is determined to be pixels, and the complexity is increased by 1 in each direction. Windows in this way Compute the complexity by comparing the color differences of all my pixels.

However, when i = M-1, the color difference is compared only in the horizontal direction, and when j = M-1, the color difference is compared only in the vertical direction. In this method, pixels having color differences that cannot be visually noticed are treated as pixels having the same color value so that they can be used as characteristic values representing complexity among image characteristics. In addition, the maximum complexity size for the M × M image block is defined as in Equation 6.

MaxC = 2 * M * M

On the other hand, the image complexity determined by the above method may be inconsistent with the quantization interval characteristic when the actual image is quantized. For example, in the image block to be encoded, most areas are composed of color-sensitive pixels and have a low complexity, but when a high complexity image appears in only one portion, the overall complexity is defined as high. A large value of quantization intervals are assigned according to the visual characteristics of being relatively insensitive, resulting in a deterioration of image quality of the reconstructed image. Accordingly, in the present invention, the input image block is divided into 4 × 4 sized dependent image blocks s (where 0 ≦ s ≦ n, n is the total number of dependent image blocks), and encoding is performed using the characteristics of the respective dependent image blocks. The above problem is solved by using a method of defining the complexity of an image block to be performed. However, (M-1) is defined as a multiple of four.

1) The number of pixels (spel (s)) generating color differences in dependent image blocks is obtained. That is, pixels having a color difference greater than pE in the horizontal or vertical direction are defined in the dependent image block.

2) The total number of dependent image blocks (sblk) satisfying spel (s) ≥3 is obtained.

3) When the dependent image block satisfying sblk≥ (n / 2) +1 or sblk = (n / 2) and spel≥3 appears as shown in FIG. 5 (c), the complexity of the image block to be encoded is If it is defined as the value defined above but does not satisfy the above conditions, the complexity is defined as 4M-2. The complexity 4M-2 is a complexity defined when one component passes diagonally through an image block composed of the same color as shown in (b) of FIG. 5, and images having a value below this complexity are classified as sensitive images.

The image complexity classified by the above method is characterized by k complexity intervals (C [k]) as shown in Equation 7, which is used to determine the allowable prediction error size according to the image characteristic in the visual perception prediction error determiner. do.

0 ≤ complexity ≤4M-2, C [0] = complexity

4M-2 <complexity ≤2 , C [complexity / 10-2] = complexity

The allowable time-difference difference adjusting unit 302 has an error that is not actually sensed visually in the prediction error image data e generated as shown in Equation (1). In addition, these errors have characteristics that vary the visual perception depending on the image characteristics. This characteristic provides the validity of having to assign the PCT weight value according to C by receiving the image complexity characteristic value C for each pixel in the image block O to be encoded from the image complexity classifier. In other words, if the complexity is high, the PCT value should be relatively high, and if it is low, it should be relatively low. Therefore, the permissible gamma color difference adjusting unit 302 defines the relationship between the image complexity and the PCT weights (WY, WU, WV) according to the complexity of the surrounding image, and uses this to define the PCT value according to the image complexity as shown in Equation (8). .

Ty = Ty * WY

Tu = Tu * WU

Tv = Tv * WV

The visual prediction error generation unit 304 predicts errors that are not visually sensed in the input prediction error image e by reducing prediction errors for Y, U, and V color components by using PCT values defined in consideration of image complexity. An error image S is generated and defined as in Equation 9.

if (eY ≦ Ty) SY = 0;

else SY = eY;

if (eU ≦ Tu) SU = 0;

else SU = eY;

if (eV ≦ Tv) SV = 0;

else SV = eY;

According to the present invention, a human visual perception chrominance characteristic is applied to each pixel to be encoded in a conventional image encoding apparatus so that the temporal correlation of image data can be improved without causing image quality deterioration for a reconstructed image. In this way, the coding efficiency of the image can be improved. On the other hand, in the present invention, the image complexity is defined using the gaze color difference characteristic, which has an advantage that can be provided in the following points more effective than the variance value mainly used to define the conventional image complexity.

1) We can consider the color characteristics that could not be defined by the Y component alone.

2) The maximum value of complexity can be known, so the degree of complexity can be known at once.

3) It is easy to classify images because the complexity range can be known.

In addition, in order to remove the prediction error, the image complexity may be taken into consideration so that more error elimination effects can be expected, thereby improving encoding efficiency. This can be compared in FIG. 6. 6A illustrates error image data, and FIG. 6B illustrates visual error image data. In addition, since the human color gamut is implemented in a look-up table, the application of the image coding method does not require an additional processing time for computation, and thus, the computation efficiency can be achieved. In addition, it has the advantage that it can be applied to maintain compatibility with the conventional video encoding apparatus.

Claims (1)

A video encoding apparatus for constructing a gaze error image S composed only of errors for generating a gaze chrominance by assigning a prediction error image e to each pixel to be encoded with a PCT weight according to image complexity, wherein the video encoding apparatus improves the temporal correlation of image data. , Instead of variance using Y component, M × M window centering on this pixel by receiving pixel O (x, y) (0≤x≤N-1, 0≤y≤N-1) of original image An image complexity classification unit for defining an image complexity of M-1 is a multiple of 4 by using a luminous color difference characteristic to give a PCT weight considering a luminous characteristic to a pixel to be encoded; The complexity defined by the image complexity classification unit is input and divided into complexity intervals of a specific size. If the complexity is large, a large PCT weight can be given otherwise or a small PCT weight can be differentially assigned according to the image complexity to encode an image to be encoded. Allowable time difference difference adjusting unit to provide a characteristic; And By comparing the PCT weights (Ty, Tu, Tv) defined by the allowable color perception difference adjusting unit with the size (ey, eu, ev) for each color component of the prediction error image e, the prediction errors smaller than the PCT weight do not have the color perception color difference. And a vision error processing unit including a vision prediction generation unit that defines a vision prediction error image S according to the visual chrominance characteristic by judging all errors by 0.