Scalable Picture And Video Coding

The technology of internet has developed very quickly since the 1960s, and so did the multimedia at the same time, so the quantity of data transmitted on the internet became much greater. The service of multimedia has become so popular in the modern time and it attracts more and more attention. For example, the services of browsing pictures on the internet, video on demand and so on, these multimedia services bring the internet one trouble after another. The data transmitting on the internet has a prodigious disadvantage, it can't afford the QoS. When the internet becomes busy it usually lost the data packages at will, because of that it impacts the effect of rebuilding, especially on the picture and video service. As the frontier of the data process, how to make the coding more effective has become urgently. So we are searching for a coding technology that can adapt the variation of the bandwidth in a certain extent. The scalable coding is just the proper one. The transition of the bitstream can also be credible when the bandwidth is limited, varied or when the channel is incredible.In the text we expound the scalable image coding and the scalable video coding, and make improvements on these coding methods. The new method has a very good result through the experiments. At first the traditional scalable image coding disparts the frequency of the picture to get the lowpass and the highpass subband of the picture. Make the lowpass subband as base layer, it has upper priority. Make the highpass subband as enhancement layer, it has lower priority. The base layer usually uses JPEG coding, the data coded will be transmitted to the internet, but this method of base layer coding doesn't exert the predominance of the scalable coding completely. On the one hand, this method has a certain request for the bandwidth, the data bandwidth that be coded must be in the lower limit of the bandwidth. When you receive part of the JPEG coding data it can't show the picture, so it will waste the data that has been received. On the other hand, this coding method hasn't adequate flexibility to satisfy the consumers'different requires. In the text we make some improvements on the scalable image coding and bring forward a classified Fine Granularity Scalable coding based on histogram.This method bring forward a new strategy mainly aims at base layer coding. It classifies the gray levels base on the probability that the pixels appear and then code the value of the pixels with different precision base on the classes they belonged. To a random picture we do twice wavelet transforms at first and then calculate the histogram of the lowpass subband of the picture and then classified the histogram. The gray levels with high probability are the first class and the gray levels with low probability are the second class. In order to find the high probability gray levels, in the text we use a method of window scanning. We use the window scan the histogram. The size of the window is set as 60% that means the ratio of the number of pixels in the window to out of it is 60%. And .the window with the minimum width is exactly the one we want to find then the gray levels be covered are set as the first class, the rest levels are set as the second class. After that we arithmetic code with the value of the pixels within the gray levels they belonged to. The value of the pixels that in the high probability levels will be coded more bits. When the entire data that be coded is decoded, the rebuild picture is the most similar one to the source picture. The value of the pixels that in the low probability levels will be distributed fewer bits. In this way we can not only control the range of the error but also make the error of the entire data decoding not very large. After the picture is coded we incise the data in the landscape orientation by bit, so we get the bit-plan data. We code the bit-plan data with the bit-plan coding and then we can transmit the coding data to the internet. The receiver receives the data bit by bit and decodes the data to show the picture. The coding and the decoding process both stop at the same time when the receiver receives the feedback that the consumers send.The most excellent character of The Fine Granularity Scalability coding is that it can control the bitstream transmitting by bit. In the other word, you can cut off the bitstream at any bit and it doesn't affect the display of the pictures. This characteristic has a prodigious predominance when the consumer wants to browse the pictures on the internet. It can stop the transition of the data when the consumer finds the picture isn't the one that he likes. So this method can save the resource. The consumer gets some information about the picture through the parts of the picture's data that received and find the region of the interest. In the text we only enhance the region of the interest of the picture. We code the region of the interest with JPEG. The region of the interest is chosen by the consumer; we call it interactive region selected. This method can use of the information of every bit to show the picture better.The scalable video coding includes temporal scalability, space scalability, frequency scalability, complex scalability, SNR scalability and so on. In the text we mainly research the temporal video scalability. To realize the temporal video scalability we usually use Motion Compensated Temporal Filter (MCTF). The odd frame in the video sequence is the prediction frame; the even frame is the reference frame. By calculating the motion estimate between the prediction and the reference frames, we get a set of motion vector. After doing the Haar wavelet transform through the direction of the motion vector, we get a lowpass frame and a highpass frame. The lowpass frame is the base layer with upper priority; and the highpass frame is the enhancement layer with lower priority. How to get more effective motion vector and how to distribute the bits are the keys to influence the effect of the MCTF. After the motion estimate, the corresponding connection between the two frames'pixels is one of the states below: one pixel in the prediction frame corresponds to one pixel in the reference frame, more than one pixels in the prediction frame corresponds to one pixel in the reference frame and the pixels in the reference frame have no corresponding pixels in the prediction frame. More correspondence to one and no correspondence is called unconnected pixels. How to code the unconnected pixels will affect the result of the video coding. As the more correspondence to one in the unconnected pixels, we use the minimum distance strategy instead of the mean value, because the motion distance of the object between the two frames is tiny. After that, there are only two states left as below: one correspondence to one and one has no correspondence pixel. The unconnected pixels in the prediction frame we call them"covered area". Because there is great probability that the area will be covered by the object when it moves from the first frame to the second frame, the unconnected pixels in the reference frame we call them"exposed area", because there is great probability that the area will be exposed when the object moves from the first frame to the second frame. In the text we save the covered pixels in the lowpass frame and save the exposed pixels in highpass frame. We select 16 sequential frames in the video sequence as a group of pictures and calculate the motion estimate between each two of them. In this way it can make 4 levels 15 times MCTF. The consumer can decode the video in different rate of frame base on the consumer's selection and the status of the internet. According to the experiment the new coding strategy increases the PSNR of the decoding picture in a certain extent.