Research On Distributed Coding For Image Array

Dense camera array has been widely used for date acquisition in many applications. Images captured from camera array often have overlapped fields of view and have large redundant data, thus data compression becomes especially important for such system. This paper focuses on the distributed coding method of camera array. We investigated the encoding method, decoding method, the correlation between adjacent nodes, the arrangement of nodes, the matching method, and the application of distributed coding in a three-dimensional measuring system in detail.In order to achieve high compression rate, the coherence among neighboring views should be exploited and the images required to be encoded jointly, however these cameras typically can not communicate with each other, and only communicate with a central node, and also the process ability of distributed encode are often very limited. So it is preferred to compress the images directly at each camera with a less complex method in a distributed manner.According to the Wyner-Ziv theorem, two correlated sources can be compressed to similar extent no matter whether they communicate with each other or not, provided that the decoder have both compressed outputs available.In the proposed method, the cameras need not to communicate with each other, while the correlations between them can still be exploited successfully, and the complexity of the encoder is very low. Part of the images are coded and decoded independently with JPEG These images are noted as K nodes and used to generate side information in the decoding process, the others nodes noted as D nodes are coded using the syndrome coding method and decoded using side information generated by K nodes. We analyzed the coding error, the number of coefficients encoded, the relationship between quantization and coset partition. For the case of low nodes correlation, a scaled quantization method was proposed, and the influence of different arrangement of camera node for the coding efficiency was discussed.In the distributed manner, the correlation between nodes is exploited in the decoding side. According to the character of camera array, we proposed the bidirectional matching method, generating high frequency in the decoding side, and matching in half-pixel resolution, and all this matching methods are proved to be effective for increasing the coding efficiency.Camera array can be used to reconstruct the three-dimensional information effectively. For the case when the scene have much depth variation, and contains a lot of occluded regions, we propose to use the image based rending method to generate a reference image for the distributed nodes using the adjacent key nodes in the decoding side and do matching. By using the distributed method, the bits need to be transmitted are saved greatly without sacrificing the reconstruct effect.