Research On Bayer CFA Imaging Processing Algorithm Based On Hybrid Model

With the high popularity of portable imaging devices, the demand of high-resolution images grows rapidly, then storage space and transmission bandwidth becomes the major factors which affect the quality of living. As Bayer filter array is one of the most popular color filter arrays, the research on Bayer image processing is a main content of information and signal technology. Bayer image needs to be interpolated in order to obtain color images. Currently, the gradient-based interpolation algorithm is applied more widely. It mainly focuses on how to use the edge or the correlation between pixels to get higher resolution color images. Therefore, research of the edge detection operator for Bayer image is of great significance. If the raw data is interpolated firstly and then compressed, it will result in asymmetry between compression and decompression resources, which is contrary to people’s demands. What’s more, the interpolated image also brings some redundancy of data and a large demand of storage space. Therefore, it’s important to study how to better compress the raw data. The main works of the thesis are presented as follows:(1) The edge detection operator for Bayer CFA image. This thesis studied several typical edge detection operators, and analyzed their efficiency for the edge detection of Bayer image. According to the special structure of the Bayer filter array, a new edge detection operator is proposed. Edge detection is done directly within the Bayer CFA image, and the algorithm has low computation. The results show it is more efficiency than other edge detection operators. It’s meaningful to Bayer image interpolation algorithm.(2) Bayer image restructuring. This thesis studied four representative Bayer image restructuring methods (separation, direct merge, weighted sampling and filtering methods). After analysis of these methods, we found that the weighted sampling method was better than others.(3) Improved SPIHT algorithm. The original SPIHT algorithm was studied in this thesis, and improved for compression of Bayer CFA image. As the original SPIHT algorithm requires a lot memory which is difficult to implement with hardware. An improved SPIHT algorithm is proposed, which is based on six special state symbols. The experimental results show the effectiveness of the algorithm. (4) Bayer image compression based on the hybrid model. The improved SPIHT algorithm achieves good performance in the low frequency region, but is not sufficient for the images which have a large number of zero-trees. A new Bayer image compression algorithm based on hybrid model has been proposed in this thesis. The hybrid model effectively combined the good performance of improved SPIHT coding in the low-frequency region and the resolution scalability property of EBCOT. After testing with different images, the proposed hybrid model shows better performance than the original method.