The Research Of Image Representation Method Based On Biological Visual Mechanism

Image is an important source for human sensory outside information, and it's also an important carrier for information communication. Explore effective methods of image representation has great significance for the research of image processes and imag depth understanding. Meanwhile, it can also improve the method of image encoding and coding, compute vision and image-based control technology. However, the conventional methods of image representation have some problems and deficiencies, such as large amount of image information data, images are easily corrupted by noise and lack of adaptability. Aimmed at these problems, an image representation model based on biological visual mechanism was proposed in this paper. Biological visual system has formed perfect information processing mechanism after a long-term evolution, and could efficiently characterize the image information. The image representation model based on biological visual mechanism was builded by simulate biological visual system process image information. The following is the main study of this paper.1. The process of image transfer in the visual pathway and mechanisms of visual information processing (receptive field properties, sparse, synchronous oscillation, etc.) are analyzed, this provided a physiological basis for the research of image representation.2. Primary visual cortex (V1 area) in the visual system is responsible for extracting the image color, shape and direction information, and plays an important role in the integration of image information. In this paper, the ICA model and TICA model were use to train image datas, and obtained basis functions consist with receptive field properties of simple cell and complex cell. This provided a foundation for efficiently characterize image information by simulate V1 area information processing mechanism.3. According to the sparsity of V1 area, this paper reseached the image representation method based on sparse response. This method could characterize image information by a small number of neurons. An image denoising algorithm based on biological visual mechanism was proposed. The algorithm was applied to image denoising experiment and achieved a good denoise effect, which verified the image representation method and denoise algorithm are valid.4. Synchronous oscillation mechanism was used to build image representation model on the foundation of images sparse representation. This model is a three layer model consists of preprocessing layer, sparse representation layer and synchronous oscillation layer. The model was applied to natural image compression experiments, and verified the effectiveness and efficiency of the image representation model based on biological visual mechanism.