KR20060109246A - Video encoding and decoding method for improving coding efficiency and apparatus thereof - Google Patents

Abstract

A moving picture encoding/decoding method and an apparatus thereof are provided to improve compression efficiency when residual macro-blocks are encoded to efficiently use media for storing or transmitting moving pictures. A method of encoding a moving picture includes a step of receiving macro-blocks constructing frames of the moving picture, a step of re-arranging pixels of the macro-blocks, a step of frequency-converting the re-arranged macro-blocks, a step of quantizing the frequency-converted macro-blocks, and a step of entropy-coding the quantized macro-blocks to output moving picture bit streams.

Description

Video encoding and decoding method for improving coding efficiency and apparatus

 In video encoding technology, compression efficiency is becoming the most important performance. The H.264 video international standard is the most highly compressed technology to date, and one macroblock is divided into 4x4 blocks to perform DCT in 4x4 blocks, and the result is adaptively used by using CABAC or CAVLC decoded data as context. Perform the encoding. This conventional technique does not achieve optimal compression efficiency in encoding residual macroblocks in motion compensation or scalable encoding techniques.

Compression efficiency is improved in encoding residual macroblocks in motion compensation or scalable coding techniques.

Abstract:

In this proposal, a method for improving SVC coding efficiency on the basis of ARR is described. The presented method proposes to implement ARR on residual signal before integer transformation in SVC enhancement layer. Experiments based on JSVM2.0 are performed with different test sequences on Palma test points. Experimental results show that the presented method gives 0.08 to 0.2dB improvement over the tested sequence based on the same bit rate. It can also be observed that the presented method effectively improves the coding efficiency of FGS enhancement layer by 14% average bit rate reduction with 0.07dB PSNR loss in enhancement layer.

ARR  for improving coding efficiency of video compression

ARR (rearrangement): The residual pixel distributions in the original 8x8 block before rearrangement and the new one after rearrangement are compared in the following figure 0.

Fig. 0 Rearrangement process.

As shown in Figure 0, the original residual block is shown as (a), after rearrangement, the residual data are shown in (b).

Inverse ARR process in the decoder, transforms the block in Fig. 0 (b) to the block in Fig. 0 (a).

ARR is useful for improving coding efficiency of residual texture signal in video coding. The residual signal whether by temporal prediction or spatial prediction, the error pixel values are most of cases close to zero. So after rearrangement, the rearranged 4x4 blocks also contain almost zero error pixel values. However, before rearrangement, the neighboring 4x4 residual blocks, the statistics of error pixel value can be different because that for example, in the case of temporal prediction using motion vectors, the neighboring 4x4 blocks will use different reference information with different motion vectors. There are video coding technologies that use already decoded neighboring blocks as context for enhancing coding efficiency of current block coding such as CABAC in JVT. ARR helps improving coding efficiency by making the statistics of neighboring residual blocks be similar to each other.

ARR  implementation for JVT  SVC

       The ARR operation on residual signal has been proposed for error resilient coding purpose. In this document, ARR is adopted for improving coding efficiency in SVC enhancement layer. In SVC, MCTF is first implemented on each spatial layer, after that, the input video signal is decomposed into high-pass and low-pass signal, then, the texture information and motion information derived from the MCTF process are coded for each low- pass and high-pass picture. The SVC encoder structure with the proposed ARR module is shown in Fig. One.

Fig. 1 example of the encoder structure using ARR

        As the figure shown, ARR module is applied to the residual texture information.

        The proposed method can also be applied to Intra texture information.

       In SVC, the input video signal of each layer after MCTF is decomposed into a set of temporal low-pass and high-pass pictures. For each picture, after MCTF, texture information and motion information are encoded, respectively. When fine grain SNR scalability (FGS) is enabled, the quality scalability is achieved by progressive SNR refinement texture coding. The presented method implements ARR on residual signal before DCT transformation in the coding process of high-pass picture for enhancement layer. The proposed method can be easily implemented with simple modification to the current JSVM coding structure. Additionally, as ARR is implemented on each 8x8 inter block where 4x4 transformation is adopted, hence only one 8x8x2 Octet memory block is required for the whole ARR process

       Above description is the implementation of the proposed method in SVC encoder. When the decoder is concerned, the inverse operation needs to be implemented for the decoding process. Corresponding to the ARR operation in encoder, the Inverse ARR is performed at decoder side to the reconstructed residual signal which is obtained from the inverse integer transform.

Experimental results

      To evaluate the performance of the proposed method, experiments based on JSVM2.0 are performed on Palma test points using the configuration files provided along with the software. In order to get the preliminary experimental results, ARR is implemented on luma residual signals. The experimental results are shown in Fig. 2 to Fig. 8.the legend name "PictureFormat, FrameRate and CodingMethod" represents an encoding method for the comparative experiment. For example, "QCIF 15 JSVM" represents QCIF picture format, frame rate of 15 frame / sec with the JSVM 2.0 encoder. The encoding method 'JSVM-ARR' represents the invented, JSVM 2.0 with ARR process.

       From the preliminary results, it can be observed that the proposed method which implements ARR in FGS coding process, achieves 0.08 to 0.2dB PSNR improvement at high bit-rate. As the proposed method is only implemented to FGS enhancement layer, so at low bit-rate, the proposed method will have similar performance as the original JSVM2.0. In addition, in order to evaluate the effectiveness of the proposed method, a simple simulation with 1 FGS layer on QCIF format base layer is made on foreman sequence. The test condition for that is illustrated in Table 1.The curves of per-frame bits reduction on FGS layer by the proposed method is given in Fig. 8.Where the ratio of bits reduction in FGS layer is obtained by (1):

r = (r _a -R) / r _a x 100 (1)

Where r is the ratio of bits reduction, minus value represents a decrease in bit rate, while positive value indicates an increase in bit rate. r _a and R represent the bits used for coding each frame by the proposed method and the original JSVM2.0, respectively.

       Experimental results in Fig. 8 clearly show that the proposed method significantly reduced the bits used for coding each frame in FGS layer. Around 14% average bit-rate reduction can be obtained by implementing the proposed method in FGS layer with the penalty of 0.07dB quality loss.

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Table 1.Test condition for analysis on FGS enhancement layer bit-rate reduction

Scalability SNR GOP Size 32 Intra period Only first frame Base layer resolution QCIF Base layer mode AVC compatible Intra mode usage On Number of FGS layers 1 (QP = 6)

Fig. 8 Foreman sequence frame by frame performance on bits reduction in FGS enhancement layer

Conclusion

Compared with the original JSVM2.0, the proposed method significantly reduced the bit rate on FGS layer by approximately 14% with ARR.

By improving the video compression efficiency, the medium is efficiently used for video storage and transmission.

Claims (9)

(a) receiving each macroblock constituting a frame of a video; (b) rearranging the pixels of the received macroblock; (c) frequency converting the relocated macroblocks; (d) quantizing the frequency converted macroblock; And (e) entropy encoding the quantized macroblock to output a video bitstream. The method of claim 1, wherein in step (a), a residual macro block by motion prediction is received. Video encoding method comprising the step of The enhancement layer predicted from the base layer in stretch coding in step (a) is performed. Receiving a residual macroblock constituting a portion comprising the It is also achieved by the luxury method. In addition, the technical problem is a video decoding method, (a) video bitst Receiving a rim; (b) Encoding the macroblock of the received video bitstream Decoding the trophy; Dequantizing the decoded macroblock; (c) constructing a macroblock by inverse frequency transforming the inverse quantized macroblock information; and (d) rearranging pixel positions of the inverse frequency-converted macroblocks to their original positions. The method of claim 4, wherein in step (e) comprises a macroblock by motion compensation, And rearranging pixel positions to original positions in a motion compensated macroblock. 5. The method of claim 4, wherein in step (e) And constructing a low block and rearranging pixel positions to the original positions in the compensated macroblock. A pixel repositioning unit for repositioning pixels of the received macroblock; A frequency converter for converting the rearranged pixels into a frequency space; A quantizer for quantizing the frequency-converted data; And an entropy encoder configured to losslessly compress the quantized data. An inverse entropy decoding unit to inversely entropy decode the received video bitstream; Prize An inverse quantization unit for inversely quantizing inverse entropy decoded data; The inverse quantization An inverse frequency converter for converting the converted data into a macroblock of a pixel space; And repositioning a pixel position of the inverse frequency-converted macroblock to an original position. Input macroblock of video and rearrange pixels effectively to hide error. The video bit stream encoded by the frequency conversion, quantization, and entropy encoding Locking is also achieved by information storage media.