Researches Of All-In-Focus Image Rendering And Saliency Detection For Light Field Imaging

In recent years, with the rapid development of the light filed imaging theories and imaging equipments, light field imaging technique has been drawing more and more attention from many researchers and institutes Moreover, people’s demands for the real visual world greatly motivate the sustaining updating of the light field imaging technique and its applications. So far, light field images have been widely used in the areas of three-dimensional scene reconstruction, field depth extraction, fast multi-view scene rendering, stereo vision matching, and etc.Up to now, the light field imaging technique has already broken through the restrictions of traditional two-dimensional photography and can effectively capture the multi-dimensional visual information of actual scenes. However, there is still a lack of good and uniform schemes to extract object and integrate information from the vast and complex data of the light field cameras. Based on the research of light field image, this paper proposes an algorithm to fuse the focus information from the light field focal stack. And on this basis, a saliency detection model is realized to extract the salient object of light field effectively.This paper mainly focuses on the researches of the light field camera Lytro 1.0 and the main contributions are listed as follows:At first, this paper systematically analyses the basic theory of the light field imaging technology. And then, based on the focal stack image sequence, we propose a focus fusion algorithm based on discrete cosine transform (DCT) and graph cut theory. This method selects the clearly focused pixels by using adaptive DCT window and iterative Graph Cut optimization, and finally gets the all-in-focus image to reconstruct all the captured focus information of the current scene. Experiments of the subjective and objective evaluations indicate that the proposed method exhibits good focus fusion performance, and clearly outperforms the existing image fusion methods on the aspect of image blocking artifacts.Besides, a kind of saliency detection model for light field is realized to extract the salient object of light field image on the basis of the obtained focal stack images and all-in-focus image. As is known to all, salient object detection is a hot topic in the field of computer vision. Compared with the traditional saliency detection methods, we use multi-images instead of single image as input to extract the salient objects. A common assumption is that the foreground always contains better object and focus information, and the background on the opposite. The focal stack from the light field camera provides rich information of focus, depth and objects, which helps to generate the background and foreground feature maps. On the basis of the extracted maps, we finally conduct the light field saliency map on the all-in-focus image. Simulations reveal that this saliency detection model can robustly detect the salient object and efficiently inhibit the background information.