KR0166728B1 - The image signal encoding apparatus for edge characteristics and the quantization level decision method - Google Patents

Abstract

According to the present invention, when an edge exists in a macro block, the macro block is divided into 16 sub-blocks to reduce the quantization level of the edge-blocked macro block, thereby eliminating the flashing phenomenon generated on the screen by the quantization error. The present invention relates to a video signal encoding apparatus and a quantization level determination method considered. As described above, the present invention divides the macroblock into 16 subblocks of 4x4 pixels when calculating the quantization level value for controlling the quantization unit in the omnidirectional analyzer and divides the minimum variation value of the activity of the subblock into the macroblock. Determined by activity. The normalized activity is calculated using the activity of the macroblock obtained here and the average activity of the frame before encoding, and the quantization level value is calculated by the normalized activity. As a result, the quantization level of the macro block in which the edge exists is lowered, thereby eliminating the glare generated on the screen.

Description

Image signal coding apparatus and quantization level determination method considering edge characteristics

1 is a control flowchart showing a method of determining a quantization level in a conventional video signal encoding apparatus.

2 is a state diagram illustrating an example in which an edge exists in a macro block, thereby increasing the quantization level.

3 is a block diagram of a video signal encoding apparatus of the present invention.

4 is a control flowchart illustrating a method of determining a quantization level in consideration of edge characteristics of the present invention.

5 is a state diagram showing an example where an edge exists in a macroblock of the present invention.

* Explanation of symbols for main parts of the drawings

3: N × N conversion section 4: Quantization section

6 buffer 7 inverse quantization unit

11: Dongchu Government 12: Ministry of Trade

13: omnidirectional analysis

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal encoding apparatus. In particular, when an edge is present in a macroblock, the macroblock is divided into 16 subblocks, thereby lowering the quantization level of the edged macroblock so as to display the image by a quantization error. The present invention relates to an image signal encoding apparatus and a quantization level determination method in consideration of edge characteristics to eliminate glare generated in an image.

Recently, in systems such as HDTV, HD VTR, digital VTR, digital camcorder, multimedia videophone, video conference, etc., video signals and audio signals are encoded and transmitted as digital signals or recorded on media, and decoded again. The reproduction method is mainly used. In such a digital video processing system, the encoding side recommends the IBBPBBPI... Code as recommended by the Moving Picture Expert Group (MPEG). If a GOP (Group of Picture) structure is present, the decoding side can perform random access processing by this GOP structure. In more detail, in MPEG, frames are classified into I (Intra), B (Bidirection), and P (Predictive) types, and they are encoded in the order of IBBPBB. I'm in order. The frames from the first I frame to just before the next I frame are called GOPs and are usually composed of 12 to 15 frames.

In order to perform encoding and decoding, a frame is divided into blocks having a predetermined size. The coding units of the frame are 8 × 8 block units, 16 × 16 macro block units, horizontal stripe slice units, and vertical rows. It is divided into column-shaped units. Conventionally, each frame is divided into macroblocks of 16x16 pixels, and the macroblocks are further divided into four blocks of 8x8 pixels (see FIG. 2). In a conventional encoding apparatus, a motion vector and a quantization level are determined in macroblock units.

1 is a control flowchart showing a method of determining a quantization level in a conventional image data encoding apparatus. Rate control and quantization control, which are recommended for encoding in MPEG, are divided into target bit allocation, rate control, and adaptive quantization. Therefore, the conventional quantization level determination method is also described separately as described above.

First, the target bit allocation process performed in step 101 will be described. First, when a specific type of frame (I, P, B) is encoded, each global complexity measure (X _i , X _p , X _b ) is calculated as follows.

Where S _i , S _p , and S _b are the total number of bits generated after encoding each frame, and Q _i , Q _p , and Q _b are the average values of the quantization levels actually used to encode all macroblocks in each frame. The target number of bits (T _i , T _p , T _b ) of the next frame in the GOP of is calculated as follows.

Here, K _p and K _b depend on a weighting matrix used for quantization, and are set to 1,0 and 1.4, respectively. R represents the number of bits remaining in encoding among the bits allocated to the GOP, and is updated to R = RS _{i, p, b} after encoding the frame. N _p and N _b are the number of P and B frames remaining after encoding in the GOP.

Second, the rate control process performed in step 102 will be described. In order to encode each macroblock j, the fullness of the buffer is calculated as follows.

Where do ⁱ , do ^p , and do ^b are the fullness of the initial buffer of each frame type, and Bj is the number of bits generated after encoding the macro block j. MB_cnt is the number of macroblocks in the frame, and dj ⁱ , dj ^p , and dj ^b are the fullness of the buffer in the macroblock j in each frame. To encode the next frame of the same type, the fullness of the final buffer is the initial fullness of the next type of buffer. The quantization level of the macroblock j is calculated as follows based on the fullness of the buffer.

Where bit_rate is the desired number of bits per second and picture_rate is the number of frames per second.

Third, the adaptive quantization process performed in steps 103 and 104 will be described. In order to calculate the spatial activity measure for the macro block j, the following is obtained from the intra pixel values.

Where Pk is the original pixel value in the 8x8 block. In the adaptive quantization process in which the quantization level is adjusted in consideration of the spatial activity, the minimum variance value of the activities of the four blocks in each macro block is obtained and determined as the activity of the macro block (step 103). The normalized activity (actj) is

Here, avg_actj is an average value of the activity j of the immediately preceding frame encoded. The normalized spatial activity N_actj has a value between 0.5-2, which depends on the activity of the macro block. When the quantization level for the macro block j is mquantj, the quantization level mquantj determined by the normalized activity N_actj is as follows (step 104).

If the quantization level is determined as described above, it is determined in step 105 whether one frame is over, and if the frame is not over, the process is repeated from the ratio control process of step 102, and if the frame is over, the process returns to step 101 Repeat from the target bit allocation process. According to the conventional quantization level determination method, even if the quantization level is set to a large value in a complex image, the human eye may not recognize the flashing caused by the increase in the quantization error.

2 illustrates an example in which edges exist in a macro block, thereby increasing the quantization level. As indicated by the dotted line, conventionally, the macroblocks composed of 16x16 pixels were divided into four subblocks composed of 8x8 pixels to calculate the activity of the macroblocks. In the case where the edge is present in the macro block even though there is not a complicated image as described above, the quantization level of the macro block is set to a large value by determining the activity value as a large value by a conventional quantization level determination method. Therefore, there was a problem that the glare generated on the screen is noticeable to the user.

The present invention is to solve the above conventional problems, an object of the present invention is to calculate the macroblock activity when calculating the macroblock by dividing the macroblock into a larger number of subblocks than conventional macroblocks existed The present invention provides an image signal encoding apparatus and a quantization level determination method in consideration of an edge characteristic to reduce glare generated on a screen by lowering a quantization level of the quantization level.

In order to achieve the above object, a feature of the present invention is a method of quantizing and encoding image data and transmitting the same through a buffer, and calculating a global complexity for a frame to be encoded and using the calculated global complexity in one GOP. Calculating the target bits of the next frame, obtaining the average activity of the previous coded frame, calculating the buffer's fullness using the target bits to encode the macroblock, and calculating the buffer's fullness. Calculating a quantization level of the macroblock by dividing the macroblock into at least a plurality of subblocks, obtaining activity of each of the subblocks, and calculating a minimum value of the activity Obtaining an activity diagram, an activity of the macroblock and an average activity of the pre-encoding frame Calculating normalized activity using the normalized activity, calculating the quantization level according to the normalized activity, and determining whether one frame is finished after calculating the quantization level, and repeating the above steps. It is in the method of decision.

In addition, another aspect of the present invention for achieving the above object is an apparatus for encoding image data in a predetermined macroblock unit, the error data obtained by subtracting a predetermined feedback data from the macroblock data of the frequency domain Transform and quantization means for quantizing the transform coefficients according to the quantization level signal determined by the omnidirectional analysis section, and the data quantized and irregularly input by the transform and quantization means are stored and transmitted at a constant speed. A buffer that outputs a signal so that no overflow or underflow occurs, and receives the fullness signal of the buffer and the macroblock data and receives the macroblock data more than four subblocks according to the buffer's fullness. Divide into blocks to determine the activity of each subblock And an omnidirectional analyzer for outputting a quantization level signal by setting the minimum value of the activity as the macroblock activity, and inverse quantization and inverse transformation processing of the quantized transform coefficients output from the transform and quantization means according to the quantization level signal. An inverse quantization and inverse transform means, a frame memory for adding the inverse transformed image data to predetermined feedback data and storing the frame data in a frame unit, and calculating a motion vector by searching the frame memory for data having a pattern most similar to the block image data. An image signal encoding apparatus in consideration of edge characteristics includes a motion estimation means and a motion compensation means for reading out block data corresponding to a motion vector of the motion estimation means from a frame memory and outputting the feedback data as feedback data.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to FIGS. 3 to 5 as follows.

3 is a block diagram of a video signal encoding apparatus to which the present invention is applied. As shown, the encoding apparatus of the present invention includes a subtractor 2 for outputting error data by subtracting predetermined feedback data from block data of a unit of a predetermined size applied through the input terminal 1. An output terminal of the subtractor 2 is connected to the N × N converter 3 to convert the error data of the subtractor 2 into data in the frequency domain. The quantizer 4 connected to the N × N converter 3 quantizes the transform coefficients of the N × N converter 3 according to the quantization level signal determined by the omnidirectional analyzer 13. The variable length encoder 5 connected to the quantizer 4 compresses the data by encoding the data quantized by the quantizer 4 into variable length. The output of the variable length encoder 5 is applied to the buffer 6 in an irregular state, and the buffer 6 stores the input data and transmits the data at a constant speed to overflow or underflow. Outputs the fullness signal SQ so that) does not occur. The fullness signal SQ of the buffer 6 is applied to the omnidirectional analyzer 13, and the omnidirectional analyzer 13 is a macroblock unit of image data and a buffer 6 applied through the input terminal 1. The quantization unit 4 receives a quantization level signal MQ of a macroblock obtained by receiving the fullness signal SQ from the subfield, and dividing the image data into 16 subblocks of 4x4 pixels according to the degree of fullness of the buffer. ) And inverse quantization unit 7 respectively. On the other hand, at the output terminal of the quantization unit 4, an inverse quantization unit 7 and an N × N inverse transform unit 8 for inverse quantization and inverse transform processing of the quantization coefficients output from the quantization unit 4, and an N × N inverse transform unit ( The adder 9 which adds the video data of 8) with predetermined feedback data is comprised. The frame memory 10 reconstructs the screen by storing the output data of the adder 9 in units of frames. The converging unit 11 connected to the frame memory 10 calculates a motion vector by finding data of a pattern most similar to the image data of the N × N block applied from the input terminal 1 from the stored data of the frame memory 10. . The compensator 12 also reads block data corresponding to the motion vector of the sync 11 from the frame data of the frame memory 10 and outputs the block data to the subtractor 2 and the adder 9. The switch SW1 connected between the compensator 12 and the subtractor 2, and the switch SW2 connected between the compensator 12 and the adder 9, have an image that is encoded due to accumulation of error data. To prevent the data from being changed, the data is refreshed in units of frames or predetermined blocks.

In the encoding apparatus of the present invention configured as described above, image data of N × N blocks is input to the input terminal 1 (generally, N ₁ × N ₂ blocks, but in the embodiment, N ₁ = N ₂ = N is assumed). The block data applied through the input terminal 1 is subtracted from the data fed back from the compensator 12 in the subtractor 2 to calculate error data. The error data is converted into data in the frequency domain by the N × N converter 3, and at this time, the energy of the converted conversion coefficient is mainly collected toward the low frequency side. The quantization unit 4 converts the transform coefficients into representative values of a predetermined level through a predetermined quantization process. That is, the quantization unit 4 is supplied with the quantization level signal MQ determined by the omnidirectional analyzer 13, thereby quantizing the output data of the N × N converter 3 by a predetermined quantization weighting matrix. . For the quantized quantization coefficient, the variable length coding unit 5 compresses the data V _CD transmitted by variable length coding utilizing the statistical characteristics of the representative values. The buffer 6 receives the compressed data supplied from the variable length encoder 5 and transmits it at a constant speed, and outputs a fullness signal SQ to adjust the amount of input data so that no overflow or underflow occurs. The omnidirectional analyzer 13 receives the fullness signal SQ and the image data in macroblock units and inverses the quantization level signal MQ determined by the fullness of the buffer 6 with the quantization unit 4. Output to the quantization unit 7.

Also, in general, since there are many similar parts between the screen and the screen, if there is a slight motion, the motion vector (MV) is calculated by estimating the motion, and when the data is compensated using the motion vector, the difference signal between adjacent screens. Is so small that it can compress the transmitted data further. In order to perform such dynamic compensation, the inverse quantization unit 7 and the N × N inverse transform unit 8 inversely quantize the quantization coefficients output from the quantization unit 4 and then inversely transform the image data in the spatial domain. The image data output from the NxN inverse transform unit 8 is added to the data returned from the compensator 12 and added by the adder 9 and stored in the frame memory 10 to reconstruct the screen.

Then, the tracking unit 11 finds the block data of the pattern most similar to the N × N block data applied from the input terminal 1 from the data stored in the frame memory 10, and shows a motion vector (MV) representing the movement between two blocks. To calculate. This motion vector (MV) is transmitted to the receiver and also to the compensator 12 for use in the decoding apparatus. The compensator 12 reads block data corresponding to a motion vector from the data of the frame memory 10 and supplies it to the subtractor 2. The subtractor 2 then calculates the error data between the block data applied at the input stage 1 and the block data supplied from the compensating section 12 as described above, and the error data is encoded and transmitted to the receiving side.

Here, the switches SW1 and SW2 are interlocked switches, which perform operations that are turned off while the macroblock is forcibly intracoated to prevent the encoded image from being different from the actual image due to accumulation of error data.

4 is a control flowchart illustrating a quantization level determination method in consideration of the edge characteristics of the present invention performed by the omnidirectional analyzer 13. First, in step 201, a target bit allocation process is performed to calculate each global complexity X _i , X _p , X _b when a specific type of frame I, P, or B is encoded. The target number of bits (T _i , T _p , T _b ) of the next frame in one GOP is calculated using the calculated global complexity, and the average value (avg_actj) of the activity of the immediately preceding frame is obtained. Next, in step 202, a rate control process is performed to calculate the degree of fullness of the buffer for encoding each macro block j. The quantization level of the macro block j is calculated using the fullness of the buffer. The target bit allocation and rate control process as described above is the same as the conventional method.

A feature of the present invention is the adaptive quantization process performed in steps 203 and 204. In this process, as shown by dotted lines in FIG. 5, the macroblock is divided into 16 subblocks of 4x4 pixels, and the minimum value of the subblock activity is determined as the macroblock activity. First, the following values are obtained from the intra pixel values in order to calculate the spatial activity for the macro block j.

Where P _k ′ is the pixel value in the 4 × 4 subblock. In the adaptive quantization process, the minimum variation value of the activities of the 16 sub blocks in each macro block is obtained and determined as the activity of the macro block (step 203). The normalized activity N_actj is calculated using the activity j of the macroblock obtained by the above equation and the average activity avg_actj of the frame before encoding. At this time, if there is actually a complicated image in the macroblock, the normalized activity is almost the same as the value obtained by the conventional method. However, as shown in FIG. 5, when there is an edge, normalized activity is obtained with a small value. Therefore, when there are edges in the macroblock, the quantization level calculated by the normalized activity (N_actj) is obtained as a low value, so that the flashing phenomenon can be reduced and the number of bits generated in encoding the entire data is not largely changed. (Step 204).

When the quantization level is determined as described above, it is determined whether one frame is finished in step 205, and if the frame is not finished, the process is repeated from the ratio control process of step 202, and when the frame is finished, the process returns to step 201. Repeat from the target bit allocation process.

As described above, the present invention lowers the quantization level of macroblocks with edges when calculating macroblock activity by dividing the macroblock into a larger number of subblocks than before. There is an effect that can be eliminated.

Claims (4)

In a method of quantizing and encoding image data and transmitting the same through a buffer, a global complexity of a frame to be encoded is calculated, a target bit number of a next frame is calculated in one GOP using the calculated global complexity, and the encoding is performed. Obtaining an average activity of the previous frame; Calculating a fullness of a buffer using the target number of bits to encode a macroblock, and calculating a quantization level of the macroblock by the calculated fullness of the buffer; Dividing the macroblock into a greater number of subblocks than at least four; Obtaining activity of each of the sub-blocks, and obtaining the minimum value of the activity as the activity of the macro block; Calculating a normalized activity using the activity of the macroblock and the average activity of the pre-encoding frame, and calculating a quantization level based on the normalized activity; And determining the end of one frame after calculating the quantization level and controlling the repeated steps. 2. The method of claim 1, wherein in the dividing into subblocks, macroblocks are divided into 16 subblocks of 4x4 pixels. The method of claim 2, wherein the activity of the macro block is calculated according to the following equation. Here, var_subblk is the activity of the subblock, and P _k 'is the pixel value in the 4x4 subblock. An apparatus for encoding image data in a predetermined macroblock unit, wherein the error data obtained by subtracting predetermined feedback data from the macroblock data is converted into data in a frequency domain, and is converted according to a quantization level signal determined by an omnidirectional analyzer. Transform and quantization means for quantizing the transform coefficients; A buffer for storing the data quantized and irregularly input by the conversion and quantization means, transmitting the data at a constant speed, and outputting a fullness signal so that no overflow or underflow occurs; Based on the buffer fullness signal and the macroblock data, the macroblock data is divided into more than four subblocks according to the buffer's fullness level, and the activity of each of the subblocks is obtained. An omnidirectional analyzer for determining the activity level of the macroblock and outputting a quantization level signal; Inverse quantization and inverse transform means for inverse quantization and inverse transform processing of the quantized transform coefficients output from the transform and quantization means according to the quantization level signal; A frame memory configured to add the inversely transformed image data to predetermined feedback data and store the information in a frame unit; Dynamic estimation means for finding data of a pattern most similar to the block image data from a frame memory and calculating a motion vector; And image compensating means for reading out block data corresponding to the motion vector of the same estimation means from the frame memory and outputting the block data as feedback data.