Research On Depth Recovery Based On Defocused Images

Information society can be regarded as image information society in modern time. In information society, computer vision is the most important part, and3D reconstruction is the most important content in computer vision, namely finds real3D information from2D images. At present,3D reconstruction technology is widely used in medical science, precision industrial detection and exploration. In image detection, because of narrow space and so on limit, the fuzzy of the image which is shooted by endoscopic imaging system is generally caused by the emergence of defocused image or motion blur. In contrast, the frequency of the occurrence of defocused image is higher, therefore, Depth from Defocus is used to obtain accurate3D visualization scene, which provides a reference for the staff.In many of DFD approaches, because of simple principle, easy realization and real-time, deterministic approach in spatial domain has been studied extensively. However, the accuracy of deterministic approache still has not high enough now. So, this paper proposes a novel approach to recovering depth from defocus, which is a deterministic approach in spatial domain and can make up for lack of precision. Around the proposed approach, the following works have been done:On the one hand, first, the formula of the novel deterministic approach to recovering depth from defocus based on moment-preserving is proposed, which establishes the relationship between the ratio of defocus radius and depth, it allows camera arbitrarily changes image distance and focal length when recovering the depth. It makes up for the traditional approach can only change a parameter when camera catches the images. Second, the image segmentation converts the gray images into the gradient images by Canny operator. Firstly, with the gray image of rice leaf as examples, application of Roberts operator, Sobel operator, Prewitt operator and Canny operator for image edge detection, the results show that the best is Canny operator, Sobel operator followed by. Compared with the traditional Sobel operator, Canny operator can detect more fine edge and have high precision. Then, convert two defocused gray images from the same scene to gradient images by using Canny operator. Finally, according to the calculation principle of moment-preserving whick makes depth recovery simple. Firstly, calculate the ratio of the area of region with large gradient value to that of the whole image region in each block for each gradient image by moment-preserving method. Secondly, calculate the ratio of one gradient image to that of the other gradient image. Last, recover depth from scene according to the ratio of the ratio.On the other hand, the simulation experiments were carried out on the novel deterministic approach to recovering depth from defocus based on moment-preserving and the traditional one respectively. The experimental results show that errors of the novel deterministic approach to recovering depth from defocus based on moment-preserving is smaller, and the average value is close to the true. In depth of400m level, the improvement of accuracy is3.67%; in depth of1000m level, the improvement of accuracy is24.25%; the improvement of average accuracy is13.96%.This paper deduces the formula of the novel deterministic approach to recovering depth from defocus based on moment-preserving, which establishes the relationship between the ratio of defocus radius and the depth, it successfully recover the depth information of the scene by using Canny operator and moment-preserving, the approach is simple in calculation, meet real-time and high precision, and it plays a useful supplement for study of DFD approaches, and has very important theoretical and practical significance.