Computer Vision Based Planar Point Positioning And Multi-Resolution Analysis

Nowadays fake print documents are everywhere, how to identify these files the same or not allows judicial identification staff confused and incomprehensible. According to the inherent texture information of paper, in this paper, we concentrate on the research of computer vision based planar point positioning in order to capture image information of paper. We proposed a method which combined different magnification with image local features to solve these issues. We first introduce the camera model, calibration methods, and complete the calibration; In order to achieve planar point positioning and visual measurements, backward projection and image feature matching were used to extracted world coordinates from a single image. To resolve similar objects, a method combined magnification with local multi-scale feature is proposed, which has been successfully applied to distinguish similar cover paper. The results showed that the proposed method has strong ability to distinguish similar objects.Firstly, we studied computer vision and MRA related technical background and applications, mainly investigated the importance of identification of print paper about judicial documents Detailed information of camera imaging model, forward projection process and camera calibration related technologies was presented, we complete the calibration based on Zhang's calibration method and analyze the calibration results theoretically and experimentally. In order to achieve planar point positioning and visual measurment, we extracte the real-world coordinates of spacial points from their image coordinates accurately, illustrate the influence of image distortion on the accuracy, experiments showed that the positioning and measurement error of the backward projection method generally did not exceed ±0.02mm.To obtain the corresponding sample points on the forensic document, planar homography estimation method based on image matching is proposed, in which we use SIFT and RANSAC to estimate the homography robustly and accomplish point positioning using the estimated homography, then we compare the results with backward projection method which indicates that the average positioning error is less than±1 pixel. For the purpose of identifying similar printed paper, learn from the thought of multi-scale expression, we put forward the identification method which combines the magnification "physical scale" and local multi-scale SURF feature of image. We did experiments and ideal results were obtained which illustrated the feasibility of this method. Experimental results show that to distinguish whether it's the same paper only 6 scales are needed.