Facial Image Compression Based On Compressed Sensing Theory

Today is a information age, people get information using computers more and more commonly, such as text, image and video etc. In recent years, the face image is used widely in the information acquisition for Large Firm employees, ID card images’ acquisition and the public security organ’s identification of authentication. Larger and larger number of digital image data leads it inconvenient to use while we storage face images. Therefore, the development of image compression technology is needed urgently. Compressed sensing has got extensive development in the field of digital image processing recently years, such as denoising, deblurring and classification of image. This article advances a new algorithm for image compressing based on the theory of compressed sensing and BCR algorithm, and proved the feasibility and the superiority of the algorithm through experiments.The main research contents and results of this article:We researched and analyzed mainstream compression algorithm of the face image, summarized its advantages and disadvantages, improved the algorithm of BCR and use it to study image compressing of face. The solution of sparse representation is faster than the OMP algorithm. In the section of quantization, due to the sparse coefficient is relatively small, we used the Sigmoid function of neural network to enlarge the number to a certain range, and then round them. At last we received the compressed stream with the help of Hoffman coding.Firstly, we compared our algorithm with BCR in the initialization of the dictionary, and saw that our algorithm was better than the combined orthogonal basis dictionary of BCR. Secondly, we compared our algorithm with JPEG, OMP facial image compression algorithm, analyzed the compression ratio, subjective quality image restoration and objective quality (PSNR values) and proved that the image can be restored better in same level of compression. Compressing facial image based on the compressed sensing improved the quality of restored image after compressing. Applying BCR algorithm instead of OMP algorithm to obtain the sparse representation of image reduced the computing time, and applying the improved initialization dictionary of this article can represent the image information better, and restore the image clearer.