KR0171747B1 - A segmentation-based encoder and decoder - Google Patents

Abstract

The present invention relates to a segment-based coder and a decoder. The segmented image is encoded by the parameter coding unit, and the received segment and the original segment are transmitted by vector quantization of the difference image signal (residual image), respectively, according to the above two methods. The video quality is improved by decoding the video after decoding them. At this time, the residual video signal is divided into N × N bands, and the overhead is reduced by encoding and transmitting only the maximum power band by comparing and calculating the power of each band.

Description

Segment-Based Encoder and Decoder

1A is a schematic diagram illustrating a typical segment based encoder.

1b is a schematic diagram illustrating a general segment based decoder.

2 is a diagram illustrating the concept of segmentation.

3 is a diagram for explaining another method of segmentation.

4 is a block diagram illustrating a segment based encoder according to the present invention.

5A and 5B are schematic diagrams for explaining band shift according to the present invention.

6 is a block diagram illustrating a segment-based decoder according to the present invention.

* Explanation of symbols for main parts of the drawings

10, 41: segmentation section 12, 41: parameter coding section

16 decoding unit 18 interpolation unit

42: video playback unit 43: subtractor

44: band pass filter group 45: maximum power band detector

46, 65: mixer 47: low pass filter (LPF)

48: down sampler 49: vector quantization unit

60: segment based decoder 62: vector quantum decoder

63: up sampler 64: low pass filter (LPF)

66: adder

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a segmentation-based encoder and decoder, and in particular, to the artificiality of a received image by encoding and transmitting a difference image (hereinafter, referred to as a residual image) between an original image and a reproduced image from encoded data. A segment-based coder and decoder that is capable of eliminating artifacts.

In general, in the conventional coding method that processes the block or macroblock unit by using the mathematical statistical characteristics of the video signal, the lower the bit rate, that is, the higher the compression rate, the blocking artifacts, mosquito artifacts and boundaries. Problems such as blurring are becoming serious and visually annoying.

Therefore, in order to reconstruct an image that satisfies the human vision, which is the subject of image quality evaluation, in a system requiring high compression ratio, the human visual system (HVS), which studies what a person sees and feels, actively uses an image signal. It is preferable to encode. These new next-generation coding schemes are becoming more active with the launch of MPEG4, the international standard, and include object-based coding, model-based coding, segmentation-based coding, and fractal coding. Second generation coding schemes are being actively studied.

Segmentation-based coding encodes by segmenting the image around the object in consideration of the fact that human vision is sensitive to the boundary in order to remove the blocking or boundary blur at low bitrates, which is a problem in the conventional video encoding techniques. The segmented image is subjected to a boundary-texture encoding that represents a boundary and an internal texture for each segmented object.

That is, as shown in FIG. 1A, the segmentation-based encoder includes a segmentation unit 10 (SEGMENTATION) for dividing an input image into a similar class according to a specific criterion, and a parameter (eg, an image of the image segmented in the segmentation unit 10). For example, it is composed of a parameter coding unit (Contour-texture coding) (12) for encoding and transmitting contours, textures, motion vectors, etc., and the segmentation-based decoder receives and decodes the image sequence encoded by the encoder. A decode section 16 and an interpolation section 18 (Intetpolation) which reproduces the decoded video by interpolation. At this time, the image segmentation method may be divided into region-based segmentation, boundary-based segmentation, and segmentation using a histogram.

First, as a representative method of region-based division, there is a split-and-merge method. As shown in FIG. 2, the image is divided into small units and then compared to similar pixels with neighboring pixels. If it is judged, it is summed, and if it is a region of different properties, it is repeated. According to this method, a relatively accurate result of image segmentation can be obtained, but a disadvantage is that the execution time is long.

Second, boundary-based segmentation is a method of tracking by boundary line, which sequentially follows the boundary of an object using the slope value and direction information of each pixel in the input image. To track. In addition, since the result is divided into a boundary line and an area surrounded by the boundary line, boundary-texture coding can be directly applied without any further processing. However, if there is noise in the input image, the preprocessing step is required because it may follow the false boundary.

Third, the method of using a histogram is the simplest and straightest method. As shown in FIG. 3, the desired object is separated from the background by setting the valley portion as a threshold by examining the brightness distribution. . The disadvantage of this method is that when there are several objects that deal with color information but have similar brightness values, it is difficult to distinguish between the objects and the division by the threshold is difficult because the valley is not clear in a complex image having various objects. In Figure 3, the axis of abscissas represents intensity, and the axis of ordinates represents accumulation.

In this way, image segmentation is the merging of regions, the divisional merging and separation of regions. The merging of regions first divides an image into basic units such as, for example, pixels, and classifies the same classes according to similarity of specific properties. The method of joining together, and the division of the domain is to repeat the process of dividing the inverse into two parts by the characteristics of a given category. The combination of merging and separating is to repeat the process of merging given neighbors if they are markedly and different, as in Figure 2, and merging similar neighbors.

The result of the segmentation is a two-dimensional region segment image, with each region given a unique label. Each region is surrounded by a closed curve, and the characteristic values of the image basic elements in the enclosed portion are within a certain range. Here, the characteristic values can be classified into spectral properties, shape properties, and borderline properties, and spectral properties include light intensity, color, and pattern shape, and shape properties include area, eccentricity, and density. The boundary properties include the degree of curvature and contrast.

However, in the conventional segmentation-based coding, after segmenting the input image according to the algorithm described above, the boundary of each segment and the values of the boundary are coded and transmitted, and the receiving side reproduces the image from this information. At this time, the receiving side makes a playback image by interpolating boundary values with linear or polynomial function values.

However, in the interpolation with the polynomial function, even if the order of the function is increased, the segmentation-based image from which the details of the image are removed has a problem that it is difficult to reduce the artificial effects as if the cartoon is drawn.

Accordingly, the present invention encodes a residual image between a segmented image and an image reproduced by interpolation, and then segments and encodes the segmented image and transmits the segmented image to improve the quality of the received image. Its purpose is to provide an encoder and a decoder.

In order to achieve the above object, the encoder of the present invention obtains a residual image by comparing an image segmented by band-spliting an input image with an image reproduced after encoding the segmented image, and extracting the residual image by n ×. After passing through n bandpass filters and dividing into n × n bands, the power of each divided band is calculated and compared with each other to select a band having the maximum power, and the selected band is low-pass converted in the mixer. The low-pass filter removes the noise, removes the noise, downsamples the low-converted residual image, and then codes the down-sampled residual image signal using a vector quantizer to transmit the first segment-coded data together with the segment-coded data. To output

In addition, the decoder of the present invention for achieving the above object, after receiving the image sequence transmitted by the encoder, decode the segment-coded data in the segment decoder, vector quantized decoding the data encoded the residual image The decoded residual image is upsampled and expanded to its original size by removing the noise with a low pass filter, and then high-pass-converted to the original band by a mixer. And the reproduced video is output.

Thus, by adding and transmitting the residual image data to the segment coded image data according to the present invention, the image reproduced from the data segment-coded and transmitted at the time of decoding and the image reproduced from the data encoded and transmitted are further combined. It is possible to obtain a playback image with improved image quality.

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

First, in the present invention, segmentation is regarded as a filtering concept.

The edge part of the segmentation has a high frequency component, and the segmentation of the image which is decoded and reproduced after segmentation has a strong low frequency component, so the difference between the image reproduced by the segmentation method and the original image is called a residual image. The spectrum of can be said to contain a lot of components of the middle band. In the present invention, in encoding the residual image as auxiliary information, a technique of analyzing a frequency component of the image and transmitting a band having strong energy is used.

Next, the present invention will be divided into an area division encoder (encoder unit) and an area division decoder (decoder unit) for convenience of understanding.

1. Division Partition Coder

The encoder according to the present invention includes a segmentation unit 40 for segmenting an input image as shown in FIG. 4 and a parameter coding unit 41 for encoding a parameter of a segmented image; An image reproducing unit 42 for reproducing an image after inputting the output of the parameter coding unit 41; A subtractor 43 for subtracting the segmented image and the image reproduced by the image reproducing unit 42 and outputting a residual image; A band pass filter group 44 (n * n BFP) for dividing the residual image of the subtractor 43 into N × N bands; A maximum power band detector 45 for calculating power for each band of the band pass filter group 44, comparing the calculated power of each band, and detecting a band having the maximum power; A mixer (46) for low-pass converting the band selected by the detector (45) from the bands passed through the band pass filter group (44); A low pass filter (LPF) for removing high frequency noise from the output of the mixer 46; A down sampler 48 for down sampling the lowpass signal; And a vector quantization encoder 49 (VQ) for quantizing and encoding the down-sampled residual image signal.

The encoder of the present invention configured as described above will be described in detail for each component block.

As described above, the segmentation unit 40 divides the image into a plurality of segments by combining the input image signals with similar signals and separating other signals, and the parameter coding unit 41 encodes and outputs the parameters of the divided image segments. do.

The image reproducing unit 42 decodes the segment-coded image sequence and outputs the reproduced segment, and the subtractor 43 subtracts the reproduced segment image from the segment image before encoding and outputs a residual image.

The band pass filter group 44 is composed of N × N band filters to divide the residual image into N × N bands, and the maximum power band detector 45 calculates power of the N × N bands, respectively. Compare with each other to detect the band with the maximum power. In other words, when N × N full bands are transmitted, the overhead due to the transmission of the residual image increases, so that only the band with the largest power is detected and then the band can be encoded and transmitted.

The mixer 46 is connected to the output of the bandpass filter group 44 to shift the maximum power band detected by the detector 45 to the low range. Multiply by

For example, a predetermined value when the hatched band is the maximum power band as shown in Fig. 5a Is multiplied by the output of the bandpass filter group 44 and moved to the low range as shown in FIG.

As described above, the maximum power band of the residual image shifted to the low pass is coded by the vector quantum encoder 49 after noise is removed from the low pass filter 47, and then output to the transmission path.

2. Region Division Decoder

As shown in FIG. 6, the region division decoder according to the present invention includes: a segmentation-based decoder 60 for inputting and decoding a segmentation-coded sequence in a received image sequence; A vector quantization decoder 62 (VQ ⁻¹ ) which inputs and decodes a sequence encoded by the residual image in the received image sequence; An upsampler 63 for upwardly sampling and expanding the decoded residual image; A low pass filter (LPF) 64 for removing high pass noise from the enlarged residual image signal; A mixer (65) for multiplying the output of the low pass filter (64) by a predetermined value to change the high range; And an adder 66 which adds the residual image converted into the high range and the segment decoded image to output a reproduced image.

The decoder of the present invention configured as described above will be described in detail for each component block.

In the video sequence encoded by the encoder according to the present invention, a segment coded bit stream and a residual stream coded by vector quantization are mixed. Therefore, the decoder according to the present invention decodes the segment coded bit stream by the segment method, and decodes the bit stream encoded by the vector quantization by the vector quantum decoder and adds the decoded video signals.

The segment-based decoder 60 of the present invention inputs and decodes a segment coded bit string in a received image sequence, and the vector quantum decoder 62 inputs and decodes a bit string in which a residual image is encoded in the received image sequence. .

The up sampler 63 enlarges the residual image signal down-sampled in the encoding process to its original size, and the low pass filter 64 removes high-frequency noise from the output signal of the up sampler 63, and the mixer 65. On the lowpass residual image Multiply by to move to the original band position. The adder 66 adds the residual image to the image decoded by the segment method and outputs a reproduced image having improved image quality.

As described above, according to the present invention, an image sequence obtained by encoding a residual image is added to the segment coded image sequence as auxiliary information, thereby improving the image quality of the image reproduced on the receiving side.

Claims (2)

A segmentation unit 40 for segmenting the input image; A parameter coding unit 41 for encoding a parameter of a segmented image; An image reproducing unit 42 for reproducing an image after inputting the output of the parameter coding unit 41; A subtractor 43 for subtracting the segmented image and the image reproduced by the image reproducing unit 42 and outputting a residual image; A band pass filter group 44 for dividing the residual image of the subtractor 43 into N × N bands; A maximum power band detector (45) for calculating power for each band of the bandpass filter group (44), comparing the calculated power of each band, and detecting a band having the maximum power; A mixer (46) for low-pass converting the band selected by the detector (45) from the bands passed through the band pass filter group (44); A low pass filter 47 for removing high frequency noise from the output of the mixer 46; A down sampler (48) for down sampling the lowpass signal; And a vector quantization encoder (49) for quantizing and encoding the down-sampled residual image signal. A segmentation-based decoder 60 which inputs a segmentation-coded sequence in the received image sequence and then decodes it; A vector quantization decoder 62 which inputs and decodes a sequence encoded by the residual image in the received image sequence; An up sampler 63 for upsampling and expanding the decoded residual image; A low pass filter (LPF) 64 for removing high pass noise from the enlarged residual image signal; A mixer (65) for multiplying the output of the low pass filter (64) by a predetermined value to perform high pass conversion; And an adder (66) for adding the residual image converted into the high range and the segment decoded image to output a reproduced image.