IT Research Paper

Digital cameras and many similar digital image sensor devices have gained significant popularity in recent years, and they are widely used in consumer and commercial areas, the archive of patent document, medical imaging, aerospace industry, national defense and security fields and so on. And the resolution of the image sensor has been more than 10 million pixels, and the transmission bandwidth and the storage space are limited. So doing the research of image compression is of great theoretical significance and of great application value.The compression technology of the raw data of Bayer image before interpolation is discussed in this paper. The main works of this are as follows:1、A thorough research of the standard of JPEG-LS is made, and the standard is implemented by programming. Typical standard test images are changed into gray scale images. And the images of gray scale are compressed using the standard, the results compression ratio of gray scale images is between 1.29-2.18. 2、Typical standard test images are changed into Bayer pattern images. And the Bayer image are used to test the standard, because the compression ratio of the Bayer image is not good, structure conversion method is introduced to improve the standard. The compression ratio of the structure conversion is between 1.17-1.6736, which indicates the validity of the method.3、Interpolation method is used before compress Bayer images through the JPEG-LS standard. That is the causal context used to predictive the current is interpolated before prediction. Experiments show that this method can achieve a higher compression ratio than that of structure conversion.4、A neural network predictive coder is designed in the paper, and the predictor can exploit the high order redundancies and the correlation of pixel between different color channels. The foundation of the neural network structure, the influence of the transfer function and the choosing of many other parameters are analyzed and discussed. Experiments show that the compression of the predictive coder is up to 1.75. Besides, the proposed predictor is simple and easy for hardware implementation, and the computational complexity of the predictor is low, so it is very suitable for real time application.5、This paper describes a novel investigation of edge detection on Bayer images. The proposed method can directly operate on the Bayer data. Experiments results show that the edge map can be retrieved successfully directly from Bayer image including some narrow edges, other benefits include the computation saving during the interpolation and the color space transformation to a full color or grayscale image, and also lower memory requirement on the same time.6、An edge detector based on the causal neighborhood pixels is proposed in this paper. The proposed edge detector generates the edge map by categorizing the Bayer pattern image into seven kinds of regions. The proposed edge detector is simple and can save computation. And the edge detector is used to improve the precision of neural network predictor; meanwhile it can also guarantee the real time character of predictor. Experiments show that the compression ratio of the neural network predictor with proposed edge detection method is between 1.21 – 1.82, which indicates the validity of the edge detection algorithm.