Study Of Incomplete Image Repairing Technology Using Tensor Analysis

In many practical problems, the images may inevitably have flaw during the image acquisition, transmission or preservation processes, such as the three-dimensional medical scanning images of pathological organs, the remote sensing images with defects, and the scratched precious images preserved under poor condition. This thesis takes those defective images as the incomplete image data.Incomplete image processing and recognition now is a new direction in the researches. One of the solutions is to make an effective repair or completion to the incomplete data, and then implement relevant processing or recognition. The current image repairing technologies can be divided into two categories. One is the geometric model based image inpainting technology, which is particularly applicable to repair the small-scale images defects. The other is the texture synthesis based image filling or completion technology, which achieves a better result in filling the images with large-area missing pieces. The focus of this paper is to take research on effective image repairing methods from incomplete dataset. Considering that tensor analysis is a very useful method to confer the geometric structure information from sparse or noised images. This thesis expects to conduct a novel incomplete data analysis and repairing method in the virtue of tensor analysis. At the first, tensor voting and its feature extraction methods would be presented.The paper also makes a few improvements to the existing tensor voting algorithm. On this basis, we propose image repairing algorithms for deletion of small regional and large regional damage respectively. In the small area repairing algorithm, we build a priority based on the tensor features and then use an iterative method to repair the image window with greatest priority. As the size of the window is automatically selected and different thresholds are applied accordingly, this method has better applicability in this field. In the large area repairing algorithm, our method including both structural and texture repairing processes, that is first to repair the missed structure property of the entire image, and then to repair the structure and texture within. Because of the inference of the missed structural information, our algorithm achieves a better visual effect and objective evaluation criterion.At last, experiments show that comparing with traditional algorithms, and our method has a better result in repairing incomplete images. Therefore, the work in this paper would extend the applicability of conventional technologies in digital image processing, computer vision and pattern recognition areas.