Research On The Approach Of HDR Vision And Its Implementation On FPGA

The world around us is so colorful that innumerable beautiful natural scenes are available everywhere, and many of them are high dynamic range (HDR) scenes. However, it is hard to reproduce a high dynamic range scene directly in a photograph or on a conventional display device, e. g. the LCD monitor. In order to handle these problems, kinds of HDR vision methods are developed. Unlike the electronic imaging devices, the human eyes can cope well with HDR scenes. By mimicking the function in light and color perception of the human visual system, Land et al. proposed the famous theory, the Retinex theory. According to Land's theory, it is supposed that the perceived lightness is closely related to the reflectance of the objects, but has no predictable relationship with the input illumination.In this paper, several categories of Retinex algorithms are analyzed and studied. Comparing their advantages and disadvantages, the McCann-Sobel algorithm, which is fit for being implemented on FPGA, is chosen and improved. And finally the improved algorithm is realized in the FPGA system.The main contributions of this paper are listed as follows:1. Several kinds of Retinex algorithms are analyzed and studied deeply in this paper;2. McCann-Sobel algorithm is improved in this paper, and in implemented in a vision processing system based on FPGA.Experiments show that the Retinex algorithm, which is implemented on FPGA in this paper, has a good performance on compressing the dynamic range of HDR photographs. By applying the improved algorithm, the influence of the illumination is removed from the original HDR photographs, and image details are enhanced with few visible artifacts.