Geometric Feature Approach Algorithms Research Of The Target Object In Digital Image

Currently, the feature extraction of the target image is one of the most important and active research topics in the image processing fields, which has wide application in practical problems of image tracking and recognition. Among the image features, geometrical characteristics are simple including perimeter, area, rectangularity, circularity, moment invariants, Fourier descriptors, etc. Although the principle and implement of the corresponding approximation algorithms for geometrical characteristics seem easy, the accuracy and adaptability of geometrical characteristics approximation are seriously limited by the fuzzication and complexity of the blurred and complicated target objects, which results in a problem that obtaining precise feature extraction of the target boundary quickly is still difficult. The research starts from the introduction of the some classic algorithms of perimeter estimation, and focuses on solving the perimeter estimation of blurred and complicated target boundary. The purpose of this paper is to improve the accuracy, robustness and speed of the perimeter estimation. To achieve this purpose, the focus of the work is on gray-level boundary, boundary tracking, image granule and the thickness of boundary and the main word includes:(1) Research of classic perimeter estimation methods. This part summarizes the current classic perimeter estimation approaches which include the method based on digital straight segment, the method based on Minimum Length Polygon and the method based on local perimeter calculating using gray-level information.(2) Research of accurate perimeter estimation of target boundary based on gray-level information. This part gives the detail on an improved method: accurate perimeter estimation based on boundary tracking and local perimeter calculating using gray-level information. The experimental results show that the proposed method is prior to the traditional approaches like DSS, MLP and GL. Meanwhile, the proposed method has the best adaptability and stability for target objects with blurred or discontinuous boundaries.(3) Research on image granule to reduce the impact of blurred and complicated boundaries on perimeter estimation of the target object. This part gives the detail on other improved method: perimeter estimation of target boundary based on image granule. Compared with the classic methods, the proposed method uses less time in the case of maintaining the accuracy of perimeter estimation.(4) Research of perimeter estimation of target boundary based on self-adapting image granule. Firstly, a probability model based on image granule is designed to define the boundary thickness of the target object; secondly, the image is preprocessed by blocking and granulating based on the optimal size of image granule which is derived from the boundary thickness; lastly, the classical algorithms are applied to the images that have been preprocessed to estimate the perimeters of the boundaries of the target objects. The experimental results show that, the proposed method can adaptively deal with different boundary thickness of the target objects, which uses less time in the case of maintaining the accuracy of perimeter estimation, and has the best accuracy and adaptability comparing with other methods, especially for images with blurred and complicated boundaries.