Research On Digital Video Enhacement Technique Based On Target

In many applications, digital video may be corrupted by high dynamic range (HDR), complex illimunation, and strong fixed pattern noise (FPN). Most of traditional enhancement algorithms based on rock-bottom appearance, such as color or gradient, neglect the content of the images, especially the targets that attracts human's attention, which lead to two aspects of conflicts in tranditional digital enhancement. One is the conflict of target's dynamic range and background's, and the other is the conflict of suppressing FPN and preserving target's edges and details.This paper proposes a novel digital video enhancement algorithm based on target, on the basic of improving the traditional enhancenmet based on spatial and temporal characteristics. In contrast to traditional enhancement algorithms, our algorithm is purposive. Firstly we divide image into saliency region and normal region using saliency detection, and then the tones of target and background are reconstructed. Finally we enhance the two regions using different algorithms respectively. As a result, the details of target are stood out, whilst the noises are suppressed. The propose method overcomes the conflicts brought by tranditional enhancement and improves the quality of digital video drastically.We test our algorithm on real and simulant image sequence, and compare it with traditional algorithms. Experiments have shown that our algorithm is more effective.This thesis systematically studies on four aspects:1. Research on HDR compression technology. In order to reject the details loss, color washed out, and halo artifact in traditional HDR compression algorithms, this paper propose a novel HDR compression algorithm based on large scale information of illimunation subsection mapping. This algorithm can compress the HDR, preserve the details, reduce the halo artifact and color distortion, and make the image more reality.2. Research on nonuniformity correction (UNC) technology. Becase of traditional UNC algorithm produce"ghost"artifact and blur the details, this paper propose a novel fixed pattern noise suppression method based on background modeling. Since FPN exists as steady states in the temporal, we can build the FPN model by background modeling. It can suppress the FPN, avoid"ghost"artifact and preserve the details of images.3. Research on video enhancement and saliency detection technology. This paper propose a novel digital video enhancement algorithm based on target. By detecting the saliency, we divide the image into two parts. One is salient region, the other is normal region. Enhance the two regions using different algorithms respectively. It can stand out the information of salient region, improve the normal region properly and make the video more useful for applications. 4. The digital video enhancement based on target are designed and realized based on CUDA and OpenMP. The efficiency of the algorithms is improved drastically.