Research In The Algorithms Of Depth From Defocusing Based On Image Processing

The algorithms of depth from defocusing (DFD) based on image processing is one of the research focuses in the field of computer vision, the DFD technique requires only two images acquired at two differet lens settings to compute depth estimates. For this technique belongs to monocular vision,it avoids the matching of the feature points existed in stereoscopic vision. Thereby this method is more practicable than stereoscopic vision in many fields.Firstly, the basic principle of the DFD and the methods of the noise removing are introduced. Secondly, this thesis expatiates several classical edge detection algorithms and a self-adaptable method of edge-detection based on the gredient magnitude, it also offers some relative analysis and comparison and has proven the self-adaptable method of edge-detection based on the gredient magnitude is one kind of better edge detection algorithm.Thirdly, two kinds of the DFD algorithms are discussed. One algorithm obtains images by regulating the imaging plane slightly and uses dynamic referencing technique which makes the result distance is regardless of object texture, but it need suppose that points spread function is the two-dimensional gaussian function and is lack of accuracy. The other method uses moment-preseving principle and estimates the distance between camera Lens and subjects through measuring image edge blurring circle, it doesn't need calculate point spread function of optical system, it is effective for the two complex objects, the varying illumination has little effect on the algorithm. The experimental result and analysis chart of the two algorithms have shown the validity of the algorithm and certain limitation.The thesis uses the median filter algorithm to remove noise, it can avoid the existing problem using the gaussian filtering to remove noise. A self-adaptable method of edge-detection based on the gredient magnitude is used in the second method ,by test validation, it proves the accurate degree of distance estimation.