Research On Image And Video Coding Based On Compressive Sensing

The compressed sensing is a new coding scheme that reduces the sampling rate at the time ofcompressing the signal.This method breaks the Nyquist sampling theorem that the sampling ratemust be equal or greater than twice times the baseband bandwidth. As the require for ultrabroadband video communication, compressed sensing will display its unique advantages in futurewireless video communication by the way of low sampling rate, low hardware requirements andhigh video reconstruction quality.This paper first describes the basic compression sample theoretical, from sparse, matrix-orientedobservation set out three compression and reconstruction algorithm of sampling the essence.Matrices for measurement, according to the measurement matrix to classify the nature, conditionsand construction methods, summed up the various construction methods and the advantages anddisadvantages of matrix. We classification the different reconstruction algorithms at the same time,and introduces the advantages and disadvantages of various types of reconstruction.Next, we analysis and research the bit-plane coding scheme. The block of image is divided intoeight bit planes in the method, and each bit plane is encoded independently. Ordinary8-bit image isdivide into eight bit-planes, the eighth plane is the highest level.In this way, according to thedifferent level of the plane,the choices of measurement vector also is different. At the same time,we divide the priority of the bit-planes and the block that the bit-plane belongs.The more importantthe bit plane has the higher priority, so,we should encode the plane at the first time. At the decodingside, we approximate or precise reconstruction every bit plane with algorithm. The differentdegrees of importance of the bit-plane for the different block of images, the program for the biterror is more robust, simple structure, easy of hardware and software applications.Then, the bit plane compression perception method was improved, the content of improved bitplane is not the bit of pixels, but a integer, i.e., the original pixel. Improved method while decodingstill need every bit plane sparse information, but improved bit-plane method, bit-plane is completeinteger coefficients set rather than the original method that bit. At this time we do not have toconsider the identifier of every bit when encode the coefficients. Recovery algorithms do not havethe additional information of bit identifier that sent to the decoder side to reconstructe the frame.Through simulation, we prove the improved bit plane coding method have superior signal-to-noiseratio and compression efficient. Finally, in order to solve the problem of redundancy existing in inter and intra frame ofcompression-aware video encoding, the paper presents an adaptive compression-aware videoencoding scheme. The program introduces the mean square error (MSE) of adjacent frames.According to the MSE, it adds reference frame in the GOP in real time to reduce the reconstructionerror of non-reference compressed perception when decoding. As different block has variousredundancy in the frame, the thesis proposes a scheme that control the sampling rate by the wayof setting the gate of different sparsity of each block. Experiments results prove that the adaptivecompression sensing video method can effectively solve the reconstruction error problems causedby dramatic scene changes and violent movement.