Video Image Preprocessing Technology Research And Implementation

Video image performance, such as contrast, dynamic range, and so on will greatly deteriorate under the foggy, cloudy, backlight, underexposed or night environments. Noise introduced from the image acquisition process of photoelectric sensor will also affect the image information acquisition and understanding. Finding the way to efficiently improve the video image contrast and reduce the impact of noise plays an important role in traffic monitoring, marine, outdoor surveillance applications. On the other hand, with the improvement of video image resolution, video image must be compressed before transmission. However, compression is based on human visual properties; large amounts of image detail information will be lost after compression and could not be recovered by decompression. So, deteriorated raw video should be adaptively enhanced and de-noised first before compression and transmission.In this paper, two aspects of pre-processing algorithms for raw video image are researched as well as their real-time implementation techniques.In terms of enhancement, the algorithm proposed in this paper combines scene classification and details preservation on the basis of histogram equalization. It divided details preserving histogram equalization into several segments according to the scene characteristics of images. Simulation results show that, the algorithm can not only avoid the swallow and over-enhancement problems of traditional histogram equalization, but also have a robust treatment effect for different scenes.In terms of de-nosing, the nonlocal means algorithm is adopted in this paper. First, the running time of original nonlocal means algorithm is successfully reduced by two orders of magnitude with lots of optimization techniques such as look-up table, weight symmetry, block non local mean and so on. Then the video images are divided into background and motion regions to de-noise respectively based on the invariance of the background in video surveillance and some other applications. In this way, the nonlocal algorithm’s working area can be reduced to achieve the optimization for real-time video de-noising in C environment.Finally, the algorithms mentioned above are realized in the TI TMS320DM648digital video development platform and the enhancement, de-nosing and real-time performance of the algorithms are also verified and tested.