Blocking 3-D DCT Based On Regional Feature And Vector Quantization For Image Coding

1. IntroductionFollowing the development of computer and communication technology, the concept of communication also has changed largely. The information that transferred through computer and other communication system is mainly the text, still picture or simple moving pictures, which can not act as intuitional, vivid, exact, high-efficiency, big-content as color video signals. The explosion of data becomes a spiny bottleneck problem that restricts the development of video technology. In order to solve this problem, we can not only depend on the advancement of hardware and the improvement of network bandwidth. Data compression is an effective method.2. Main research work in this paperThis paper puts forward an algorithms called color image coding based on variable size 3-D DCT and VQ through studing the technology of 3-D DCT(3-Dimensioned Discrete Cosine Transform) ,VQ (Vector Quantization) and many algorithms for image compression invented by other persons.This method not only can take the advantage of correlation of three frames(R/G/B) but also can decrease the spacial redundancy of contiguous block.Then the distribution of transformed coefficient can have many traits and we can use vector quantization to compress the data.In this algorithm,3-D DCT and VQ play a key role.Because many algorithms invented before take the three frames(R/G/B) of a 8×8 image block as a whole and transform them in the same time with 3-D DCT.It can decrease correlation between three frames than before.But it can not decrease the special redundancy between contiguous image block in the frame,so both 2-D DCT and 3-D DCT have a major drawback that the decoded images, especially at very low bit rate and coping with images containing lots of high frequency,exhibit highly noticeable blocking artifacts near the block boundaries.Many other algorithms divide the image into 8×8 blocks,then classify them and assemble the congener block into a 8×8×8 cube.We see the cube as a whole,transform it with 3-D DCT.This method thinks about the correlation within the frame only and it fails to decrease the redundancy of the three frames(R/G/B).So its effect is not evident.Accordingly,this paper puts forward a new method.It can decrease both the correlation between 3 fames(R/G/B) and the redundancy of the contiguous image block.It is called color image coding based on adaptive block-size 3-D DCT and VQ.Simulation results demonstrate that it can improve compression ratio and be able to reconstruct image with content effect.The main idea of the algorithms is described as follows.First the color image is transformed into gray.Then transform it with 2-D DCT.Calculate the feature coefficient sum(described in detail in the paper).The feature coefficient sum can well reflect how much high frequence in this image block.Depending on this sum,we can classify the image block(8×8) three grades.In the table every cell corresponds to a 8×8×3 image block.Secondly,we scan the table.If the feature coefficient sum meets 0≤S Y≤ε1,we note it as"0";if the figure meets S Y>ε2,we note it as"2";other instances we note as"1".So we can construct a new table called quantization table.Thirdly transform the image block depending on the quantization table.If there are four"0"snuggle each other,it means to transform the image block with 16×16×3 3-D DCT(four 8×8 equal 16×16);"1"means to transform the image block with 8×8×3 3-D DCT;"2"means to transform the image block with 4×4×3 3-D DCT(divide the block into 4 4×4×3 blocks).We can understand only when the four 8×8 image blocks share the similar feature coefficient sum,they can be transformed with 16×16×3 3-D DCT as a whole;Only when the four 8×8 image blocks have little correlation with each other, they can be transformed with 4×4×3 3-D DCT;Other instances are transformed with 8×8×3 3-D DCT.Thus we not only can decrease the correlation of three frames(R/G/B) but also can decrease the spacial redundancy of contiguous image block.The quantization table must be transmitted inerrably because the table is necessary for decoding to reconstruct the image.When the vector quantizer works the transforming mode table also plays an important role.Depending on the table the quantizer can know which mode is using now.Scanning the coefficient vector,quantizer chooses the corresponding codebook to get the nearest codeword.Because the algorithms studied in this paper have a relative complex operation,so it is not suitable for real time image process.This also allows us to design the codebook and the fast codeword search method in the vector quantization more easily because of thinking no complex problem.This will also offer us with the least distortion and the biggest compression ratio.Then image quality could be assured.So this paper has designed three codebooks,corresponding to the different transforming mode.Also we enlarge the dimension of the vector to achieve bigger compression ratio.Weighted error measure is adopted to ensure that lower frequence has less distortion.