Research Of Color Correction Method For Multiview Video

Multiview video is obtained by applying different cameras to take videos of different viewpoints of a particular scene simultaneously, thus getting information on different aspects. However, a problem arises with the increased emplacement of cameras; the amount of data also increases exponentially alongside with camera numbers and therefore leads to further problems in data compression, data transmission and arbitrary viewpoint rendering. Inter-view prediction is a key method in solving the problems mentioned above. As videos taken down by co-operating cameras varies in many factors such as position, illumination rate of scenes, allocation of cameras, CCD noise, shutter speed, exposure rate etc. Correlation will therefore be sacrificed between different viewpoints, leading to errors in inter-view prediction. Hence, we have to acquire an efficient color correction method to solve non-consistence color appearance in both luminance and chrominance between images of different viewpoint, thus increasing the accuracy in inter-view prediction and the coding efficiency of multiview video processing.We firstly introduced the multiview video system as well as its key technologies in this paper, followed by a thorough introduction of the contemporary research status and main methods of color correction. The fixed-block matching algorithm is often used in the current color correction methods, but it is acquired by disparity estimation and therefore applying the same disparity vector to every individual pixel in the block. If a particular block consists of various objects, or it is located in the boundaries, shaded or exposed region, then the disparity vector would create an unignorable error in the reliability of visual estimation if the size of the blocks were to vary. In this paper, a variable-size block matching based color correction method for multiview video is proposed. Firstly the matching blocks is acquired by disparity estimation, and then classify their reliability according to the residual energy. Secondly, a quadtree segmentation method is adopted to correct and refine the unreliable disparity vectors with variable-size blocks, so the visual artifacts due to incorrect disparity vectors can be eliminated and the object boundaries and structural information can be preserved. Finally, color correction coefficient is obtained by linear regression method to correct the color difference between different viewpoints, and histogram equalization is used to further improve the calibration quality, providing both better subjective and objective quality.