KR20100027612A - Image compressing apparatus of lossless and lossy type - Google Patents

Abstract

PURPOSE: An image lossless/losing compression apparatus is provided to facilitate a chip implementation by lowering the complexity through a DG decoding and a space prediction device. CONSTITUTION: An image lossless/losing compression apparatus comprises the following: a space prediction device(130) inputted with a gradation data regarding a perimeter pixel involved to a specific pixel from a block butter(110); a subtractor(150) subtracting a prediction data regarding the from the specific pixel from the space prediction device, and a graded data inputted from the block buffer; and a decoder(190) adjusting a bit rate regarding a lossless error data by a decoding parameter inputted from a parameter calculator(170).

Description

IMAGE COMPRESSING APPARATUS OF LOSSLESS AND LOSSY TYPE}

The present invention relates to an image compression device, and more particularly, to an image lossless and lossy compression device for reducing memory bandwidth, which can efficiently reduce a high memory bandwidth generated by a large amount of image data through a simple compression structure.

Chips that perform real-time image processing on an image having a size larger than HD (High Definition) level require a high memory bandwidth only by writing / reading original image data. Since the mid-1990s, frame compression technology has been researched and developed to reduce the memory (SDRAM) size of HD MPEG decoder chips. As simple hardware-based compression schemes, adaptive differential pulse coded modulation (ADPCM) schemes or down conversion schemes have been proposed.

In recent years, video processor chips not only handle HD, Full HD, or higher-quality video, but also feature a variety of video processing features, such as MPEG, 3D graphics rendering, and frame-rate conversion, which require higher memory bandwidth. do.

To solve this problem, existing methods use multiple memory controllers, expensive chip packages or many data interfaces. However, since these methods increase the cost of the compression system, the memory bandwidth reduction technique through lossless / lossy compression of images has emerged as an important core technology of image and product differentiation.

The digital image compression method is divided into a lossless compression method and a lossy compression technique, which compresses image data by removing spatial, temporal and stochastic redundancy. Image compression standardization is largely divided into the JPEG (Joint Photographic Experts Group) standardization group related to still image compression and the MPEG (Moving Picture Experts Group) standardization group related to video compression.

Research and development of the still image compressor method has been mainly carried out by the JPEG standardization group, and recently, lossless compressor methods as well as lossless compressor methods have been proposed through standardization of JPEG-2000, JPEG-LS, and the like.

Still image compression methods use block-by-block DCT transform or wavelet transform and achieve high compression ratio by using predictive coding, quantization, and entropy coding, while hardware complexity is very high.

For example, the JPEG-LS method is a lossless compression method using relatively simple predictive coding and entropy coding. However, hardware complexity is relatively low compared to JPEG and JPEG-2000, but still has high complexity. JPEG-LS consists of DPCM, run-length coders, and context modeling-based Glob Rice (GR) coders.

Such JPEG-LS operates in two modes (normal mode and run mode) by distinguishing whether or not there is little change in the image. In the normal mode, to obtain more accurate prediction data, the prediction data is obtained by using a fixed predictor and an adaptive proof sheet, and then modified GR (Golomb Rice) code is used. In run mode, the interrupt method (coding the run length and the interrupt sample value) and the consumption method (coding the run length only) are used.

Among other lossless compression methods, there is a PNG (Potable Network Graphics) method with a good compression ratio.

PNG is a file that replaces GIF. The structure of the file is the same as GIF, so you can use the advantages of GIF such as transparency effect and it has better compression rate than GIF. This method turns the original image into a PNG image and creates a data stream through the encoder. In this process, the alpha samples are separated and indexed to convert each sample to the correct value by the index of the palette, to generate the seventh compressed image from the first reduced image, and to filter type. Compression and chunks are created in order to create bitstreams.

This PNG compression method is also very inefficient because it has a structurally high complexity like the JPEG-LS method.

An object of the present invention is to provide an image lossless and lossy compression device for reducing memory bandwidth that can efficiently reduce the high memory bandwidth caused by a large amount of image data through a simple compression structure.

Another object of the present invention is to compress and store a large amount of image data without loss of quality with a simple compression structure, to efficiently reduce the memory bandwidth and to reduce the image bandwidth to reduce the memory bandwidth that can improve the compression speed lossless and lossy compression To provide a device.

An image lossless compression apparatus of the present invention for achieving the above object comprises: a block buffer in which gradation data of pixels are stored in macroblock units in which an image is divided into predetermined rows and columns; A spatial predictor that receives grayscale data of at least one neighboring pixel related to a specific pixel to be predicted from the block buffer and predicts the grayscale data of the specific pixel using at least one direction correlation between the neighboring pixels; A subtractor for subtracting grayscale data of a specific pixel input from the block buffer and prediction data for the specific pixel input from a spatial predictor and outputting subtracted error data; And an encoder for adjusting the number of bits according to encoding parameters inputted from a predetermined parameter calculator after the lossless error data inputted from the subtractor and delivering the adjusted bitstream to the memory side.

In detail, the parameter calculator may compare the correlation values of at least one or more directions between neighboring pixels input from the spatial predictor, calculate the number of data bits of the minimum correlation value as a result of the comparison, and determine the encoding parameter as an encoding parameter. The spatial predictor predicts grayscale data using a correlation between at least one neighboring pixel among pixels located in a horizontal direction, a vertical direction, a left diagonal direction, and a right diagonal direction based on a specific pixel to be predicted. It is done.

In addition, the spatial predictor receives gray level data of neighboring pixels from a block buffer based on a specific pixel to be predicted, and mutually subtracts the neighboring pixels in horizontal and vertical directions to obtain horizontal and vertical correlation data, respectively. Subtractor; A first comparator for comparing horizontal and vertical correlation data inputted from the subtractor and outputting a selection control signal according to a comparison result; A pixel operator which receives grayscale data of neighboring pixels from a block buffer based on a specific pixel to be predicted, and adds pixels located on the left and top sides and subtracts pixels located in a diagonal direction; A second comparator comparing the absolute value of the operation result value input from the pixel operator with a preset reference value and outputting a selection control signal according to the comparison result; A first multiplexer for selectively outputting grayscale data of a left pixel and a calculation result value input from a pixel operator according to a selection control signal of the second comparator; A second multiplexer for selectively outputting grayscale data of the upper pixel of the specific pixel to be predicted according to the selection control signal of the second comparator and an operation result value input from the pixel operator; And a third multiplexer for selectively outputting grayscale data or calculation result values of the left and top pixels input from the first and second multiplexers according to the selection control signal of the first comparator.

The spatial predictor may further include a clipping unit configured to clip the predicted value output from the third multiplexer to a maximum value or a minimum value within a set range when the predicted value output from the third multiplexer is out of a preset reference value.

According to an aspect of the present invention, there is provided an image loss compression apparatus including: a block buffer in which gradation data of pixels are stored in a macroblock unit in which an image is divided into predetermined rows and columns; A line buffer in which error data according to data loss processing of each pixel is lost-compensated within a predetermined range and the loss-compensated compensation data is stored; A spatial predictor that receives compensation data for at least one neighboring pixel related to a specific pixel to be predicted from the line buffer and predicts grayscale data for the particular pixel using at least one direction correlation between the neighboring pixels; A subtractor for subtracting grayscale data of a specific pixel input from the block buffer and prediction data for the specific pixel input from a spatial predictor and outputting subtracted error data; A loss processor which processes the error data of the specific pixel inputted from the subtractor according to a loss factor inputted from a predetermined loss factor selector and outputs the lost error data; And an encoder for adjusting the number of bits according to encoding parameters inputted from a predetermined parameter calculator and transmitting the adjusted bitstream to the memory side.

Specifically, the loss factor selector counts and accumulates the number of bits of each pixel output through the encoder, and calculates a loss factor to be applied to the current macroblock by using the accumulated number of bits of the previous block and the target number of blocks per block. The loss factor selector counts the number of bits of error data of all pixels output from the encoder and accumulates the number of bits in each slice unit including the macroblock and the LMB unit including the macroblock. ; And calculating surplus bits for the current slice or LMB based on the target bit amount of the pre-stored macroblock, and then comparing the calculated current slice or LMB with the surplus bits and the set reference value. And a loss factor calculator for determining a factor.

In addition, the loss factor calculator is characterized by reducing the loss factor when the block change degree is low by comparing the data on the input block activity (activity) with a preset reference value.

The parameter calculator may be configured to compare correlation values of at least one or more directions between neighboring pixels input from the spatial predictor with each other, and to calculate the number of data bits of the minimum correlation value as a coding parameter.

The spatial predictor predicts grayscale data based on a correlation between at least one peripheral pixel among pixels located in a horizontal direction, a vertical direction, a left diagonal direction, and a right diagonal direction based on a specific pixel to be predicted. .

The spatial predictor is a subtractor that receives error compensation data for peripheral pixels based on a specific pixel to be predicted from a line buffer, and subtracts each other between horizontal and vertical peripheral pixels, and then acquires horizontal and vertical correlation data, respectively. ; A first comparator for comparing horizontal and vertical correlation data inputted from the subtractor and outputting a selection control signal according to a comparison result; A pixel operation unit which receives error compensation data of peripheral pixels based on a specific pixel to be predicted from a line buffer, adds pixels located at the left and top sides and subtracts pixels located at a diagonal direction; A second comparator comparing the absolute value of the operation result value input from the pixel operator with a preset reference value and outputting a selection control signal according to the comparison result; A first multiplexer for selectively outputting error compensation data of a left pixel with respect to a specific pixel to be predicted according to a selection control signal of the second comparator and an operation result value input from a pixel operator; A second multiplexer for selectively outputting error compensation data of the upper pixel with respect to a specific pixel to be predicted according to the selection control signal of the second comparator and an operation result value input from the pixel operator; And a third multiplexer for selectively outputting error compensation data or calculation result values of left and top pixels input from the first and second multiplexers according to a selection control signal of the first comparator.

The spatial predictor may further include a clipping unit configured to clip the predicted value output from the third multiplexer to a maximum value or a minimum value within a set range when the predicted value output from the third multiplexer is out of a preset reference value.

The compression apparatus may perform a left shift of the lost error data inputted from the loss processor to recover the lost data, and add the recovered data and the predictive data input from the spatial predictor to output the error compensated data to the line buffer. It characterized in that it further comprises a restorer, the loss processor is characterized in that the shifter for the loss processing by shifting the error data to the right (light) according to the loss factor.

As described above, according to the present invention, the compression device is composed of a simple spatial predictor and a GR encoder to operate in units of blocks, which is suitable for processing a large amount of image data and reducing memory bandwidth, and implements a chip due to the low complexity of the compression device. In addition to this ease, there is an advantage in terms of data processing speed and cost.

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

1 is a functional block diagram illustrating an image lossless compression system according to an embodiment of the present invention, in which a compression system includes a compression device 100 for lossless compression of an image, and data compressed through the compression device 100. Memory 300, a memory controller 350 for mutual data interface between the memory 300 and an external device, and a main controller 400 for controlling the operations of the various devices.

The lossless compression apparatus 100 compresses an image in macroblock units when compressing an image. In this case, the lossless compression apparatus 100 selectively receives information about an image size or a target compression ratio from a main controller 400 through a local bus and processes the image.

The compression device 100 includes a block buffer 110, a spatial predictor 130, a subtractor 150, a parameter calculator 170, and an encoder 190.

The block buffer 110 temporarily stores an image in macroblock units in which the entire original image is divided into predetermined rows and columns (M × N). The macroblock (MB) may be generally composed of 16 × 16 pixels and the like, and each pixel may be composed of 8 bits of gradation data. May be composed of 32x32, 8x8, 32x16, or 16x8 pixels, and the number of data bits of each pixel may also be variously added or subtracted.

The spatial predictor 130 receives gray level data of at least one neighboring pixel (a, b, c) related to a specific pixel (i) to be predicted from the block buffer 110 and correlates horizontal and vertical relations between the neighboring pixels. After predicting the grayscale data for the specific pixel using the output the predicted data to the subtractor 150. Here, when the prediction by the spatial predictor 130 is perfect, the predicted gradation data and the original gradation data are the same, and in this case, the output value of the subtractor becomes '0', thereby improving the compression ratio. However, since the gradation data of a specific pixel is predicted using the correlation of the neighboring pixels of the specific pixel to be predicted, it is not easy to make a perfect prediction that the output value of the subtractor becomes '0' for all the pixels whose gradation changes.

The subtractor 150 subtracts gradation data of the specific pixel i input from the block buffer 110 and prediction data P for the specific pixel input from the spatial predictor 130. The subtracted error data e is output. The peripheral pixels correspond to the previous pixels as pixels positioned in the left, top, and left diagonal directions with respect to a specific pixel.

The parameter calculator 170 receives the correlation values in the vertical and horizontal directions between the neighboring pixels input from the spatial predictor 130, compares them with each other, calculates the number of data bits of the minimum correlation value as a coding parameter, and determines the encoding parameter. .

The encoder 190 adjusts the number of bits according to the coding parameters input from the parameter calculator 170, and transmits the adjusted bitstream to the memory 300 side after the lossless error data input from the subtractor 150 is adjusted. In addition, the encoder 190 applied to the present invention may be a GR encoder (Golomb Rice Coder) for adjusting the number of bits according to the k value, which is a GR (Golomb Rice) parameter.

As such, instead of storing the grayscale data of the characteristic pixel in the memory 300 as it is, the grayscale data is predicted after predicting the grayscale data for the specific pixel by using the correlation between the neighboring pixels of the specific pixel to be predicted by the direction prediction algorithm. Error data is obtained by subtracting the data from the original gradation data of a specific pixel, and the obtained error data is stored in a memory by adjusting the number of bits according to an encoding parameter.

In addition, the space predictor 130 is configured as shown in FIG.

That is, the spatial predictor 130 includes a first subtractor 131 and a second subtractor 132, a first comparator 133, a pixel operator 134, a second comparator 135, and a plurality of multiplexers 136a to 136c. And a clipping unit 137.

The first and second subtractors 131 and 132 receive gray level data of neighboring pixels a, b, and c from the block buffer 110 based on a specific pixel i to be predicted, and are horizontal and vertical. It is configured to obtain horizontal and vertical correlation data after subtracting each other between neighboring pixels in the direction. In detail, the first subtractor 131 is configured to obtain a left pixel a and a left pixel for a specific pixel i to be predicted. The gray level data of the diagonal pixel c are respectively inputted and subtracted from the block buffer 110, and the subtracted vertical correlation data dy is output, and the second subtractor 132 has an upper end for the specific pixel i to be predicted. The grayscale data of the pixel b and the left diagonal pixel c are respectively received from the block buffer 110 and subtracted, and the subtracted horizontal correlation data dx is output.

The first comparator 133 compares the absolute values of the horizontal and vertical correlation data dx and dy input from the first and second subtractors 131 and 132, respectively, and then selects a selection control signal according to the comparison result. )

The pixel operator 134 receives gradation data of neighboring pixels from the block buffer 110 based on a specific pixel to be predicted, and the pixels a and b located at the left and the top are added to each other and positioned diagonally. Pixel c is subtracted.

The second comparator 135 compares the absolute value of the operation result value input from the pixel operator 134 with a preset reference value and outputs a selection control signal according to the comparison result. The reference value set in the second comparator is preferably 255.

The first multiplexer 136a selectively selects the grayscale data of the left pixel a and the operation result value input from the pixel operator 134 according to the selection control signal of the second comparator 135. Output

The second multiplexer 136b selectively outputs grayscale data of the upper pixel for the specific pixel to be predicted according to the selection control signal of the second comparator 135 and an operation result value input from the pixel operator 134.

The third multiplexer 136c selects the gray level data or the calculation result of the left and top pixels a and b inputted from the first and second multiplexers 136a and 136b, and selects a control signal of the first comparator 133. Optionally output according to

If the value of the data input from the third multiplexer 136c is outside the preset reference range, the clipping unit 137 clips the output to the maximum value or the minimum value of the set range.

That is, the spatial predictor 130 predicts prediction data of a specific pixel to be predicted as shown in Equation 1 below.

If | dy | <| dx | and (| b + dy |)> 255, Prediction Data = b

              If (| b + dy |) ≤255, Prediction Data = b + dy

If | dy |> | dx | and (| a + dx |)> 255, Prediction Data = a

              If (| a + dx |) ≤255, Prediction Data = a + dx

Here, dx is b-c, dy is a-c, a is data of a pixel located to the left of a specific pixel to be predicted, b is data of a pixel located to the top of a specific pixel to be predicted, and c is a specific to be predicted. Data of the pixel located on the left diagonal of the pixel.

In the case of such a lossless compression algorithm, a given memory bandwidth may be exceeded due to the limitation of compression when intra-block correlation is low. Therefore, a lossless compression scheme as shown in FIG. 2 and a lossy compression scheme as shown in FIG. Should be applied.

3 is a functional block diagram illustrating an image loss compression system according to another embodiment of the present invention, wherein the compression system includes a compression device 200 for lossy compression of an image by a predetermined bit amount, and data compressed through the compression device. It includes a memory 300 is stored, a memory controller 350 for mutual data interface between the memory 300 and an external device, and a main controller 400 for controlling the operation of the various devices.

The compression apparatus 200 compresses the image in macroblock units when compressing the image. In this case, the compression apparatus 200 receives information about the size of the image, the target compression ratio, and the like through the local bus from the main controller 400 to process the image.

The loss compression apparatus 200 includes a block buffer 210, a line buffer 220, a space predictor 230, a subtractor 240, a loss processor 250, a loss factor selector 260, and a parameter calculator ( 270, an encoder 280, and a pixel restorer 290.

The block buffer 210 temporarily stores an image in macroblock units in which the entire original image is divided into predetermined rows and columns (M × N). The macro block may be generally composed of 16 × 16 pixels and the like, and each pixel may be composed of 8 bits of gradation data. 32x32, 8x8, 32x16 or 16x8 pixels or the like, and the number of data bits of each pixel can also be variously added or subtracted.

The line buffer 220 temporarily stores the compensation data in which the loss error data of each pixel output from the loss processor 250 is loss-compensated within a predetermined range for each pixel. That is, in the block buffer 210, original grayscale data c_d1, b_d1, a_d1, and i_d1 of the pixels c, b, a, and i are stored as shown in FIG. 4A, but in the line buffer 220, FIG. As in 4b, loss-compensated compensation data c_d2, b_d2, a_d2, and i_d2 for each of the pixels c, b, a, and i are stored. The gray level data c_d1, b_d1, a_d1, and i_d1 of the pixels and the loss compensation data c_d2, b_d2, a_d2, and i_d2 of each pixel have the same number of bits, but data may be different due to loss. .

The spatial predictor 230 receives error compensation data of at least one peripheral pixel (a, b, c) related to a specific pixel (i) input to the subtractor 240 from the line buffer 220 and is horizontal between the peripheral pixels. After predicting the grayscale data for a specific pixel using correlation such as vertical, the predicted grayscale data is output to the subtractor 240. Here, when the prediction by the space predictor 230 is perfect, the predicted gradation data and the original gradation data are the same, and in this case, the output value of the subtractor becomes '0', thereby improving the compression ratio. However, since the gradation data of a specific pixel is predicted by using the correlation of the neighboring pixels of the specific pixel to be predicted, it is not easy to make a perfect prediction that the output value of the subtractor becomes '0' for all the pixels whose gradation changes.

The subtractor 240 includes gradation data of a specific pixel i input from the block buffer 210 and prediction data (p) related to the specific pixel i input from the spatial predictor 230. And subtracts the subtracted error data (e). The peripheral pixels correspond to previous pixels with pixels positioned in the left, top, and diagonal directions with respect to a specific pixel, respectively.

The loss processor 250 processes the error data of a specific pixel input from the subtractor 240 according to a loss factor input from the loss factor selector 260 and then outputs the lost error data. The loss processor 250 is a shifter to shift the error data to the right according to the loss factor to remove the least significant bit (LSB) to reduce the error value. For example, when the input error data is '1001' and the loss factor is 1, the loss processor 250 shifts the error data 1 bit to the right to output '100'. In other words, if the loss factor is 1, the error data is shifted 1 bit to the right to reduce the error value by 2 times. If the loss factor is 2, the error data is shifted to the right 2 bits to reduce the error value by 4 times and the loss factor is 3 In this case, the error value is shifted three bits to the right to reduce the error value by eight times. That is, according to the loss factor, the error value is reduced by 2 ⁿ (n is the loss factor value).

The loss factor selector 260 counts and accumulates the number of bits of each pixel output through the encoder 280 and applies the current macroblock using the accumulated number of bits of the previous block and target bits of each block. This is determined by calculating the lossy factor to do.

The parameter calculator 270 receives the correlation values in the vertical and horizontal directions between the neighboring pixels of the specific pixel to be predicted from the spatial predictor 230, compares them with each other, and calculates and encodes the data bits of the minimum correlation value. Determined by parameters.

The encoder 280 adjusts the number of bits according to the encoding parameter input from the parameter calculator 270 after the loss error data input from the loss processor 250 and transmits the compressed bitstream of the adjusted specific pixel to the memory side. do. In addition, the encoder 280 applied to the present invention may be a GR encoder (Golomb Rice Coder) for adjusting the number of bits according to the k value, which is a GR (Golomb Rice) parameter.

In addition, the pixel restorer 290 includes a shifter and an adder (not shown). The pixel restorer 290 recovers the least significant bits lost by left shifting the lost error data input from the loss processor 250. Error compensation data is obtained by adding the loss compensation data and the prediction data p input from the spatial predictor 230. The pixel restorer 290 outputs the error compensation data of the specific pixel thus obtained to the line buffer 220. In addition, since the pixel restorer 290 recovers only the number of bits for the least significant bit lost, it does not accurately restore the data (0 or 1) to the least significant bit lost. Pixel data stored in the line buffer 220 may be different.

As described above, the lossy compression apparatus 200 does not store the grayscale data of the specific pixel to be predicted in the memory 300 as it is but uses the correlation between the neighboring pixels of the specific pixel to be predicted according to the direction prediction algorithm. After predicting the gradation data, the gradation data is subtracted from the original gradation data of a specific pixel to obtain error data, and the loss data is stored in memory by compressing the obtained error data appropriately according to the target compression ratio set according to the loss factor. . Of course, the error data is stored in the memory by adjusting the number of bits in accordance with the coding parameters.

The space predictor 230 is configured as shown in FIG. 5.

That is, the spatial predictor 230 includes a first subtractor 231, a second subtractor 232, a first comparator 233, a pixel operator 234, a second comparator 235, and a plurality of multiplexers 236a to 236c. And a clipping unit 237.

The first subtractor 231 receives the error compensation data of the left pixel and the left diagonal pixel with respect to the specific pixel to be predicted from the line buffer 220 and outputs the subtracted vertical correlation data. For example, the first subtractor 231 receives a, which is a left pixel, and a c, which is a left diagonal pixel, for a specific pixel, respectively, from the line buffer 220, and then subtracts the vertical correlation data (dy = a-c). Will print).

The second subtractor 232 receives the error compensation data of the top pixel and the left diagonal pixel for the specific pixel to be predicted from the line buffer 220 and outputs the subtracted horizontal correlation data. For example, the second subtractor 232 receives b, which is a top pixel for a specific pixel i, and c, which is a left diagonal pixel, respectively, and subtracts it, and then outputs subtracted horizontal correlation data (dx = b-c).

The first comparator 233 compares the absolute values of the vertical and horizontal correlation data respectively input from the first subtractor 231 and the second subtractor 232, and then outputs a selection control signal according to the comparison result. The error compensation data of each pixel stored in the line buffer 220 may have a negative value according to a situation, and in this case, an absolute value of each pixel is input to the first and second subtractors 231 and 232, respectively. .

The pixel operator 234 receives the error compensation data for the neighboring pixels related to the specific pixel to be predicted from the line buffer 220 and adds or subtracts them according to a set algorithm, and then outputs the calculated result. For example, the calculation algorithm adds the error compensation data a_d2 and b_d2 of the left pixel a and the top pixel b of the specific pixel i to be predicted to each other and adds the error of the left diagonal pixel c to the added value. The compensation data c_d2 is further subtracted.

The second comparator 235 compares the absolute value of the operation result value input from the pixel operator 234 with a preset reference value and outputs a selection control signal according to the comparison result. It is preferable to set the reference value to a maximum value (2 ^n- 1, n is the number of bits) on the basis of the gray scale data (n bits). For example, since the gradation data of each pixel is 8 bits, it can have a maximum value of 255 (2 ^8-1 ) in decimal, so that the reference value (2 ⁿ -1; n is the number of bits of the gradation data) is set to 255.

The first multiplexer 236a receives error compensation data of the left pixel a with respect to a specific pixel to be predicted, and an operation result value output from the pixel operator 234, respectively, to the selection control signal of the second comparator 235. Optionally output accordingly.

The second multiplexer 236b receives the error compensation data of the upper pixel b for the specific pixel to be predicted and the operation result value output from the pixel operator 234, respectively, and receives the selected control signal of the second comparator 235. Optionally output accordingly.

The third multiplexer 236c receives error compensation data or calculation result values of the left and top pixels a and b output from the first and second multiplexers 236a and 236b, respectively, and selects the first comparator 233. Output selectively according to the control signal.

The clipping unit 237 outputs the predicted data by clipping to a maximum value or a minimum value of the set range when the value of the data input from the third multiplexer 236c is outside the preset reference range. For example, when the reference range of the clipping unit 237 is set to 0 to 255, if the value input from the third multiplexer 236c is 256, it is clipped to the maximum value of 255 and output.

In the spatial predictor 230 configured as described above, in order to predict the gray scale data of a specific pixel, the horizontal and vertical correlations between neighboring pixels with respect to a specific pixel are used. Will be calculated using Correlation between the horizontal and vertical directions is obtained using the gradient between horizontal pixels (dx = b-c) and the gradient between vertical pixels (dy = a-c) through the first subtractor 231 and the second subtractor 232. At this time, the prediction data output from the third multiplexer 236c according to the correlation is as follows.

That is, when the comparison result of the first comparator 233 is | dy | <| dx | and the comparison result of the second comparator 235 is (| b + dy |)> 255, the prediction data output from the third multiplexer 236c is b-pixel error compensation data (Prediction Data = b), the comparison result of the first comparator 233 is | dy | <| dx | and the comparison result of the second comparator 235 is (| b + dy |) <255. The prediction data output from the third multiplexer 236c is a result value (Prediction Data = b + dy) obtained by subtracting the error compensation data of the pixel c and the error compensation data of the pixel b and the pixel a. Further, when the comparison result of the first comparator 233 is | dy |> | dx | and the comparison result of the second comparator 235 is (| a + dx |)> 255, the prediction data output from the third multiplexer 236c is If a pixel is error compensation data (Prediction Data = a), the comparison result of the first comparator 233 is | dy |> | dx | and the comparison result of the second comparator 235 is (| a + dx |) <255. The prediction data output from the third multiplexer 236c is a result value (Prediction Data = a + dx) obtained by subtracting the error compensation data of the pixel c and the error compensation data of the pixel a and the pixel b.

By comparing the vertical correlation value | dy | and the horizontal correlation value | dx |, as in the above condition, it can be seen whether the pixel to be predicted is more correlated in either the horizontal direction or the vertical direction. If the value of dx is smaller, the error compensation data of b pixels is closer to the gradation data of the specific pixel i to be predicted. If the value of dy is smaller, the error compensation data of a pixel is the specific pixel to be predicted. It is closer to the gradation data of (i). If the calculated value a + b-c of a pixel, b pixel and c pixel is 255 or less, the arithmetic result value of a + b-c is used as the predictive data p.

On the other hand, if the image is actually compressed in the macroblock unit, there are cases where there are no pixels to refer to. In this case, direction prediction is impossible. The pixels in this case are predicted using differential pulse coded modulation (DPCM) to increase the compression ratio. That is, the first pixel of the block is unpredictable, but it is preferable to predict the first row in the horizontal direction and the first row in the vertical direction by using each previous pixel. For example, when only the error compensation data for the c and b pixels, which are the first rows of the macroblock, are stored in the line buffer 220, if the pixel is to be predicted, the error compensation data of the c pixel is input to the subtractor 240. If only the error compensation data for the c and a pixels, which are the first columns of the macroblock, are stored in the line buffer 220, the error compensation data of the c pixels is input to the subtractor 240 if the pixel is to be predicted. It becomes prediction data.

In the case of FIG. 5 configured as described above, since only horizontal and vertical directions of pixels to be predicted are basically considered, prediction performance may be slightly reduced when the correlation between the pixels in the diagonal direction is high. In order to secure this, as shown in FIG. 6, the prediction is performed considering the diagonal direction as well as the horizontal and vertical directions.

6 is a functional block diagram showing a detailed configuration of the spatial predictor according to another embodiment of the present invention. The spatial predictor 230 may include a first subtractor 231, a second subtractor 232, and a comparator ( 233, a plurality of multiplexers 234a and 234b, an adder 235, a shifter 236, a third multiplexer 237, a third subtractor 238, and a clipping unit 239.

In FIG. 6, the first subtractor 231 outputs error compensation data for pixels in horizontal, vertical, and diagonal directions that are highly correlated between a specific pixel to be predicted according to a preset algorithm and a line buffer 220. Each of the pixels having a high correlation is subtracted from each other and outputted by subtracting the correlation data. For example, the first subtractor 231 includes a plurality of subtractors 231a to 231h, each subtractor 231a to 231h having a high correlation with respect to horizontal, vertical, left diagonal, and right diagonal directions based on a specific pixel. Each pixel is subtracted from each other to output subtracted correlation data.

The second subtractor 232 receives and subtracts absolute values of correlation data of a plurality of highly correlated horizontal, vertical, left and right diagonal lines from the first subtractor 231, and then subtracts the horizontal, vertical, and left diagonal lines. And correlation data for the preferential diagonal line, respectively. For example, the second subtractor 232 includes a plurality of subtractors 232a to 232d, and each subtractor 232a to 232d includes a plurality of horizontal, vertical, left diagonal, and right outputs from the first subtractor 231. Among the correlation data in the diagonal direction, the correlation data in the same direction is subtracted from each other, and the correlation data for the subtracted horizontal, vertical, left diagonal, and right angle lines are respectively output. The correlation data output from the first subtraction unit 231 may have a negative value according to a situation, and in this case, an absolute value of each correlation data is input to the second subtraction unit 232.

The comparator 233 compares the correlation data for the horizontal, vertical, left diagonal and right angle lines input from the second subtraction unit 232 and outputs a selection control signal according to the comparison result.

The first multiplexer 234a receives correlation data for horizontal, vertical, left and right diagonal lines from the first subtractor 231 on the basis of the previous pixel a of the specific pixel i, and the comparator 233. According to the selection control signal of the) any one of the input correlation data is selected and output. The previous pixel of the specific pixel i is the pixel a adjacent to the left of the specific pixel i and corresponds to the pixel a immediately preceding the specific pixel i.

The second multiplexer 234b receives correlation data for horizontal, vertical, left diagonal, and right diagonal lines from the first subtractor 231 based on the previous pixel n of the specific pixel i, and the comparator 233. According to the selection control signal of the) any one of the input correlation data is selected and output. The previous pixel of the specific pixel i is a pixel n spaced apart from the left of the specific pixel i and corresponds to the previous pixel n of the pixel a adjacent to the left of the specific pixel i.

The adder 235 adds and outputs correlation data input from the first multiplexer 234a and the second multiplexer 234b, respectively.

The shifter 236 removes the least significant bit by shifting the data input from the adder 235 by one bit to the right.

The third multiplexer 237 receives error compensation data a, b, c, and f for the left, top, left diagonal, and right diagonal lines from the line buffer 220, respectively, based on a specific pixel to be predicted. According to the selection control signal of 233, any one of the input compensation data is selected and output.

The third subtractor 238 subtracts the compensation data input from the third multiplexer 237 and the shifter 236, respectively, and outputs the subtracted compensation data.

When the value of the data input from the third subtractor 238 is out of the preset reference range, the clipping unit 239 is clipped to the maximum value or the minimum value of the set range and output to the subtractor 240. For example, when the reference range of the clipping unit 239 is set to 0 to 255, when the value input from the third subtractor 238 is 256, the clipping is output at 255, which is the maximum value.

In the spatial predictor 230 configured as shown in FIG. 6, when the correlation between the pixels in the diagonal direction is higher than the pixels between the horizontal and the vertical pixels only when the horizontal and vertical directions are considered, the prediction performance may be slightly reduced. The gradation data of a specific pixel is predicted by considering the right diagonal direction.

That is, the correlation of each direction is predicted through the first subtractor 231 and the second subtractor 232 as shown in Equation 2 below.

horizontal = | △ cb |-| △ na |

vertical = | △ dn | － | △ ca |

left diagonal = | △ en | － | △ da |

right diagonal = | △ cn | － | △ ba |

Here,? Cb = c-b,? Na = n-a,? Dn = d-n,? Ca = c-a,? En = e-n,? Da = d-a, ? Cn = c-n,? Ba = b-a, a is compensation data of a pixel located to the left of a specific pixel to be predicted, n is compensation data of a pixel located to the left of a pixel, and b is to be predicted Data of the pixel located on the top of a specific pixel, c is the compensation data of the pixel located to the left of b pixels, d is the compensation data of the pixel located to the left of c pixels, e is the compensation data of the pixel located to the left of the d pixels, f Is compensation data of a pixel located to the right of b pixels.

As shown in Equation 2, two gradient values for each direction were used to increase the prediction probability of the direction.

Accordingly, the comparator 233 compares and calculates correlation data for four directions including horizontal and vertical, left diagonal, and right diagonal, selects the direction having the smallest value, and selects a control signal for the selected direction. Output to the multiplexers 234a 234b and 237. The smallest correlation value is the closest to the gradation data of the specific pixel i to be predicted.

Subsequently, since the smallest correlation value is the closest to the gradation data of the specific pixel i to be predicted, the spatial predictor 230 calculates as shown in Equation 3 below to add weights to the error compensation data in the direction. Predict the prediction data more accurately.

horizontal = a- (△ cb + △ na) / 2

vertical (vertical) = b- (△ dn + △ ca) / 2

Left diagonal = c- (△ en + △ da) / 2

Right diagonal = f- (△ cn + △ ba) / 2

That is, when the comparator 233 first outputs a control signal for selecting a direction having the smallest correlation value for each direction to the first multiplexer 234a and the second multiplexer 234b, the first multiplexer as shown in Equation (2). 234a and the second multiplexer 234b respectively output the correlation values (gradients) related to the directions to the adder 235, and the adders 235 add the respective correlation values to the shifter 236. . The shifter 236 may right shift the value input from the adder 235, that is, reduce the input value by 1/2 and output the result to the third subtractor 238.

Meanwhile, the third multiplexer 237 outputs error compensation data for pixels in a predetermined direction proximate to a specific pixel to be predicted according to the selection control signal input from the comparator 233 to the third subtractor 238. Accordingly, the third subtractor 238 subtracts data input from the third multiplexer 237 and the shifter 236 and outputs the data to the clipping unit 239.

The clipping unit 239 may limit the prediction data input from the third subtractor 238 to a preset range of 0 to 255 and then output the subtractor 240.

In the case of actually compressing an image, since there is no pixel to refer to because it is compressed in macroblock units, in this case, direction prediction is impossible. The pixels in this case are predicted using differential pulse coded modulation (DPCM) to increase the compression ratio. That is, the first pixel of the block is unpredictable, but it is preferable to predict the first row in the horizontal direction and the first row in the vertical direction by using each previous pixel.

On the other hand, in order to increase the compression efficiency in the compression apparatus of FIG. Accordingly, the parameter calculator 270 receives the data output from the shifter 236 of the spatial predictor 230, calculates the number of bits, and then determines the calculated number of bits as the encoding parameter k value and transmits the calculated number to the encoder 280. Done.

That is, the data Pe output from the shifter 236 of the spatial predictor 230 is represented by Equation 4 below in accordance with each direction (horizontal, vertical, left diagonal and right diagonal), and the parameter calculator 270 is the above. The number of bits of the data Pe output by Equation 4 is checked and determined as an encoding parameter.

Horizontal: Pe = (△ cb + △ na) / 2

Vertical: Pe = (△ dn + △ ca) / 2

Left Diagonal: Pe = (△ en + △ da) / 2

Right Diagonal: Pe = (△ cn + △ ba) / 2

FIG. 7 is a flowchart illustrating a process of determining a loss factor for one macroblock in the loss factor selector 260. Referring to FIG.

According to the type of the image processing chip, an image may be written and read in units of slices or large macroblocks (LMBs). In some cases, both may be necessary. The slice and the large macroblock (LMB) are shown in FIG. 8. For example, one row of macroblocks (MBs) in an image may be one slice, 16 macroblocks may be one large macroblock, and one slice may be included in a plurality of large macroblocks.

Therefore, from a chip perspective, it is necessary to make the memory bandwidth constant in units of slice and LMB. 7 is a proposed algorithm for adjusting the loss factor in units of macroblocks to fit a given amount of compressed bits. In other words, all the pixels in the macroblock have the same loss factor.

First, the bit number counter 262 of the loss factor selector 260 counts the number of bits of error data of all pixels output from the encoder 280. The count value is a slice unit including each macro block and the It counts and accumulates in LMB units including macroblocks (S1). That is, the bit number counter 262 may have one slice counter and a plurality of LMB counters including the slice separately.

In such a state, the loss factor calculator 263 calculates surplus bits for the macroblock, the slice, or the LMB based on the target bits of the macroblocks stored in the target bit storage 261. Each is calculated (S2).

The number of excess bits is a value indicating how many bits (occurrence bits) generated up to the previous macroblock (MB) are smaller than the number of bits (target bits) allocated to the previous macroblocks. This means that the grayscale data of the macroblock should be further compressed by increasing the factor.

The current number of surplus bits for the slice and LMB is obtained from Equation 5 below.

Current LMB_Residual Bits = Previous LMB_Residual Bits + (MB_Number of Target Bits-MB_Number of Generated Bits)

Current slice_surplus bits = previous slice_surplus bits + (MB_target bits-MB_number of generated bits)

MB_target bits of Equation 5 are values obtained by converting the bandwidth limitations of the slice and the LMB into macroblocks, respectively.

The reason for considering the number of excess bits of the slice or LMB in determining the loss factor is to increase the compression efficiency and the reliability of the image quality. For example, when the number of target bits of the macroblock is set to 100, when the number of generated bits of the first macroblock is 90 and the number of generated bits of the second macroblock is 110 in the same slice, if only the number of target bits of the macroblock is considered, The macroblock should determine the loss factor such that the number of generated bits is 90 by the number of generated bits of the second macroblock. However, in the present invention, since the number of surplus bits of the first and second macroblocks is 0 (10+ (100-110)) according to the number of surplus bits in the slice, the number of bits generated in the third macroblock is the target number of bits. The loss factor can be set to 100. This method can prevent deterioration in image quality by forcibly compressing the macroblock according to the target number of bits.

Therefore, the loss factor calculator 263 checks the number of surplus bits of the slice or LMB to which the current macroblock is obtained as described above to compare and determine whether it is equal to, smaller than, or larger than zero (S3).

If the number of surplus bits in the current slice or LMB is less than or equal to zero, the compression factor is increased by increasing the loss factor of the next macroblock by one bit than the current loss factor (S4), and the number of surplus bits in the current slice or LMB is zero. If larger, the compression factor is reduced by reducing the loss factor of the next macroblock by one bit than the current loss factor (S5).

In addition, the loss factor calculator 263 compares the data on the block activity inputted from the gradient calculator 265 with a preset reference value to determine that the block change degree is small and changes. If the degree data is larger than the reference value, it is determined that the block degree of change is high (S6, S7).

의 식에 의해 구해질 수 있다.In the above, the gradient calculator 265 calculates a gradient through correlation of neighboring pixels or macroblocks based on a specific pixel or macroblock. For example, assuming that there are a, b, c, and d pixels on the basis of a specific pixel i, respectively, the degree of change is

Can be obtained by consciousness.

Subsequently, the loss factor calculator 263 adjusts the loss factor according to the calculated block variability. If the change factor is low, the loss factor calculator 263 reduces the compression factor by reducing the loss factor (S8). This is necessary to reduce the blocking artifacts when loss occurs in the flat area.

Finally, the loss factor calculator 263 clips the obtained compression loss factor within a range of a predetermined minimum value min_LP and a maximum value max_LP, and then determines the clipped loss factor as a loss factor of the next macroblock. (S10). In the above case, when the pixel in the block has low correlation or the target bit number is set too small, the limit memory bandwidth may not be satisfied. In this case, setting the maximum loss factor (max_LP) to prevent such a case may be prevented. can do.

The compression apparatus configured as described above can reduce the processing time and hardware complexity compared to the algorithm such as JPEG-LS or PNG, and can achieve bandwidth reduction with almost no loss of image quality. In addition, when such a compression device is applied to image processing chips in various application fields such as various home appliances, portable media devices, and digital broadcasting, it is possible to obtain great effects such as price competitiveness enhancement and differentiation technology.

Preferred embodiments of the present invention are disclosed for purposes of illustration, and those skilled in the art will be able to make various modifications, changes, and additions within the spirit of the present invention. Therefore, such modifications, changes and additions should be determined not only by the claims below, but also by equivalents to those claims.

1 is a functional block diagram illustrating an image lossless compression system according to an embodiment of the present invention.

2 is a detailed functional block diagram illustrating the spatial predictor of FIG. 1 according to the present invention.

3 is a functional block diagram illustrating an image loss compression system according to another embodiment of the present invention.

4A and 4B are diagrams for explaining data stored in the block buffer and the line buffer of FIG. 3.

5 is a detailed functional block diagram illustrating the spatial predictor of FIG. 3 according to an embodiment of the present invention.

6 is a detailed functional block diagram of the spatial predictor of FIG. 3 according to another embodiment of the present invention.

FIG. 7 is a flowchart illustrating a process of the loss factor selector of FIG. 3 obtaining a loss factor.

8 is a diagram illustrating macroblocks and slice and large macroblocks of an image.

* Explanation of symbols for the main parts of the drawings

100: lossless compression device 110: block buffer

130: space predictor 150: subtractor

170: parameter calculator 190: encoder

200: lossy compression device 210: block buffer

220: line buffer 230: space predictor

240: subtractor 250: loss handler

260: loss factor selector 261: target beater storage unit

262: bit count counter 263: loss factor calculator

265: gradient calculator 270: parameter calculator

280: encoder 290: pixel restorer

Claims (22)

A block buffer in which grayscale data of pixels are stored in macroblock units in which an image is divided into predetermined rows and columns; A spatial predictor that receives grayscale data of at least one neighboring pixel related to a specific pixel to be predicted from the block buffer and predicts the grayscale data of the specific pixel using at least one direction correlation between the neighboring pixels; A subtractor for subtracting grayscale data of a specific pixel input from the block buffer and prediction data for the specific pixel input from a spatial predictor and outputting subtracted error data; And And an encoder which adjusts the number of bits according to encoding parameters input from a predetermined parameter calculator and transmits the adjusted bitstream to the memory side, after the lossless error data inputted from the subtractor is adjusted. The method according to claim 1, The parameter calculator compares correlation values of at least one or more directions between neighboring pixels input from the spatial predictor, calculates the number of data bits of the minimum correlation value as a result of the comparison, and determines an encoding parameter as an image lossless method. Compression device. The method according to claim 1 or 2, The spatial predictor predicts grayscale data using a correlation between at least one peripheral pixel among pixels located in a horizontal direction, a vertical direction, a left diagonal direction, and a right diagonal direction based on a specific pixel to be predicted. Image lossless compression device. The method according to claim 1 or 2, And the spatial predictor predicts prediction data of a specific pixel to be predicted using Equation 1 below. Equation 1 If | dy | <| dx | and (| b + dy |)> 255, Prediction Data = b               If (| b + dy |) ≤255, Prediction Data = b + dy If | dy |> | dx | and (| a + dx |)> 255, Prediction Data = a               If (| a + dx |) ≤255, Prediction Data = a + dx Where dx is b-c, dy is a-c, a is data of a pixel located on the left of a specific pixel to be predicted, b is data of a pixel located on the top of a specific pixel to be predicted, and c is a specific data to be predicted. Data of the pixel located on the left diagonal of the pixel. The method according to claim 1 or 2, The space predictor, A subtractor which receives gradation data of neighboring pixels from a block buffer based on a specific pixel to be predicted, mutually subtracts the neighboring pixels in horizontal and vertical directions, and obtains horizontal and vertical correlation data, respectively; A first comparator for comparing horizontal and vertical correlation data inputted from the subtractor and outputting a selection control signal according to a comparison result; A pixel operator which receives grayscale data of neighboring pixels from a block buffer based on a specific pixel to be predicted, and adds pixels located on the left and top sides and subtracts pixels located in a diagonal direction; A second comparator comparing the absolute value of the operation result value input from the pixel operator with a preset reference value and outputting a selection control signal according to the comparison result; A first multiplexer for selectively outputting grayscale data of a left pixel and a calculation result value input from a pixel operator according to a selection control signal of the second comparator; A second multiplexer for selectively outputting grayscale data of the upper pixel for the specific pixel to be predicted according to the selection control signal of the second comparator and an operation result value input from the pixel operator; And And a third multiplexer for selectively outputting grayscale data or calculation result values of the left and top pixels inputted from the first and second multiplexers according to a selection control signal of the first comparator. Device. The method according to claim 5, The spatial predictor may further include a clipping unit configured to clip the prediction value output from the third multiplexer to a maximum value or a minimum value within a set range when the predicted value output from the third multiplexer is out of a preset reference value. The method according to claim 5, The subtractor may include: a first subtractor configured to subtract grayscale data of a left pixel and a left diagonal pixel of a specific pixel to be predicted from the block buffer and output subtracted vertical correlation data; And a second subtractor which receives the gray level data of the upper pixel and the left diagonal pixel of the specific pixel to be predicted from the block buffer, respectively, and subtracts the gray level data from the block buffer and outputs the subtracted horizontal correlation data. A block buffer that stores gradation data of pixels in macroblock units in which an image is divided into predetermined rows and columns; A line buffer in which error data according to data loss processing of each pixel is lost-compensated within a predetermined range and the loss-compensated compensation data is stored; A spatial predictor that receives compensation data for at least one neighboring pixel related to a specific pixel to be predicted from the line buffer and predicts grayscale data for the particular pixel using at least one direction correlation between the neighboring pixels; A subtractor for subtracting grayscale data of a specific pixel input from the block buffer and prediction data for the specific pixel input from a spatial predictor and outputting subtracted error data; A loss processor which processes the error data of the specific pixel inputted from the subtractor according to a loss factor inputted from a predetermined loss factor selector and outputs the lost error data; And And an encoder for adjusting the number of bits according to coding parameters input from a parameter calculator and transmitting the adjusted bitstream to the memory side, after the loss error data inputted from the loss processor is adjusted. The method according to claim 8, The loss factor selector counts and accumulates the number of bits of each pixel output through the encoder, and calculates and determines a loss factor to be applied to the current macroblock by using the accumulated number of bits of the previous block and the target number of blocks per block. An image loss compression device. The method according to claim 8 or 9, The loss factor selector may include: a bit number counter that counts the number of bits of error data of all pixels output from the encoder and accumulates the slice unit including each macro block and the LMB unit including the macro block; And calculating surplus bits for the current slice or LMB based on the target bit amount of the pre-stored macroblock using Equation 2 below, and calculating the surplus bits of the current slice or LMB. And a loss factor calculator for comparing the set reference values with each other to determine a loss factor. Equation 2 Current LMB_Residual Bits = Previous LMB_Residual Bits + (MB_Number of Target Bits-MB_Number of Generated Bits) Current slice_surplus bits = previous slice_surplus bits + (MB_target bits-MB_number of generated bits) However, slice is a set of macroblocks (MBs) in a row in the image, and a large macroblock (LMB) is a set of macroblocks. The method according to claim 10, And a loss factor calculator to reduce the loss factor when the block variation is low by comparing the data on the input block activity with a preset reference value. The method according to claim 8 or 9, The parameter calculator compares the correlation values of at least one or more directions between neighboring pixels input from the spatial predictor, calculates the number of data bits of the minimum correlation value as a result of the comparison, and determines the image loss as an encoding parameter. Compression device. The method according to claim 8 or 9, The spatial predictor predicts grayscale data using a correlation between at least one peripheral pixel among pixels located in a horizontal direction, a vertical direction, a left diagonal direction, and a right diagonal direction based on a specific pixel to be predicted. Image Loss Compression Device. The method according to claim 8 or 9, And the spatial predictor predicts prediction data of a specific pixel to be predicted using Equation 3 below. Equation 3 If | dy | <| dx | and (| b + dy |)> 255, Prediction Data = b               If (| b + dy |) ≤255, Prediction Data = b + dy If | dy |> | dx | and (| a + dx |)> 255, Prediction Data = a               If (| a + dx |) ≤255, Prediction Data = a + dx Where dx is b-c, dy is a-c, a is data of a pixel located on the left of a specific pixel to be predicted, b is data of a pixel located on the top of a specific pixel to be predicted, and c is a specific data to be predicted. Data of the pixel located on the left diagonal of the pixel. The method according to claim 8 or 9, The space predictor, A subtractor which receives error compensation data on peripheral pixels based on a specific pixel to be predicted from a line buffer, and subtracts mutually between peripheral pixels in a horizontal and vertical direction and then obtains horizontal and vertical correlation data, respectively; A first comparator for comparing horizontal and vertical correlation data inputted from the subtractor and outputting a selection control signal according to a comparison result; A pixel operation unit which receives error compensation data of peripheral pixels based on a specific pixel to be predicted from a line buffer, adds pixels located at the left and top sides and subtracts pixels located at a diagonal direction; A second comparator comparing the absolute value of the operation result value input from the pixel operator with a preset reference value and outputting a selection control signal according to the comparison result; A first multiplexer for selectively outputting error compensation data of a left pixel with respect to a specific pixel to be predicted according to a selection control signal of the second comparator and an operation result value input from a pixel operator; A second multiplexer for selectively outputting error compensation data of the upper pixel with respect to a specific pixel to be predicted according to the selection control signal of the second comparator and an operation result value input from the pixel operator; And And a third multiplexer for selectively outputting error compensation data or calculation result values of the left and top pixels inputted from the first and second multiplexers according to a selection control signal of the first comparator. Compression device. The method according to claim 15, The spatial predictor may further include a clipping unit configured to clip the predicted value output from the third multiplexer to a maximum value or a minimum value within a set range when the predicted value output from the third multiplexer is out of a preset reference value. The method according to claim 15, The subtractor may include: a first subtractor configured to subtract error compensation data of a left pixel and a left diagonal pixel from a line buffer for a specific pixel to be predicted, and output subtracted vertical correlation data; And a second subtractor configured to receive and subtract error compensation data of the upper and left diagonal pixels of the specific pixel to be predicted from the line buffer, and output the subtracted horizontal correlation data, respectively. . The method according to claim 8 or 9, And a pixel restorer for restoring the lost data by left shifting the lost error data inputted from the loss processor, and adding the recovered data and the predictive data inputted from the spatial predictor to output error compensated data to the line buffer. Image loss compression device. The method according to claim 8, And the loss processor is a shifter which shifts the error data to the right according to a loss factor to perform loss processing. The method according to claim 8 or 9, The spatial predictor is to predict the correlation between the pixels located in the horizontal and vertical, left diagonal and right diagonal direction of the specific pixel to be predicted using the following equation (4). Equation 4 horizontal = | △ cb |-| △ na | vertical = | △ dn | － | △ ca | left diagonal = | △ en | － | △ da | right diagonal = | △ cn | － | △ ba | However,? Cb = c-b,? Na = n-a,? Dn = d-n,? Ca = c-a,? En = e-n,? Da = d-a,? Cn = c-n Δba = b-a, a is compensation data of a pixel located to the left of a specific pixel to be predicted, n is compensation data of a pixel located to the left of a pixel, and b is a pixel located at the top of a specific pixel to be predicted. Where c is the compensation data of the pixel located to the left of b pixels, d is the compensation data of the pixel located to the left of c pixels, e is the compensation data of the pixel located to the left of d pixels, and f is located to the right of b pixels. Pixel compensation data. The method of claim 20, The spatial predictor selects a direction having the smallest correlation value according to Equation 4, and calculates error compensation data of the selected direction using Equation 5 below to predict prediction data of a specific pixel. Image loss compression apparatus, characterized in that to obtain. Equation 5 horizontal = a- (△ cb + △ na) / 2 vertical (vertical) = b- (△ dn + △ ca) / 2 left diagonal = c- (△ en + △ da) / 2 right diagonal = f- (△ cn + △ ba) / 2 The method according to claim 21, The spatial predictor compresses and outputs the obtained prediction data by clipping to a maximum value or a minimum value within a set range when the obtained prediction data deviates from a preset reference value.

Images