Research And Application Of Adolescent Originality Evaluation Model Based On Image Similarity Calculation

In the selection of talented people, it’s a very important step in the process of selection to assess the learners’creativity through originality assessment. Currently, the general way of originality assessment is that testees draw a series of graphics according to the requirements then the testers, based on the image template library, evaluate the graphics by visual contrast and give the scores correspondingly. However, when the image template library and graphics are relatively large, such method is not only time-consuming but also easy to give a false score. Therefore, how to efficiently and accurately evaluating the learners’originality through graphics becomes an urgent problem to be solved. In this thesis, an originality assessment model based on image similarity calculation is proposed. It firstly extracts the shape of the target and corners of edge, and then describes the shape by Edge Gradient Direction Histogram (EDH). Finally, it marks all the graphics according to the Earth Mover’s Distance (EMD). The thesis mainly researches on how to improve the accuracy of shape extraction.The main work and innovations are as follows:(1) In order to deal with the defects of edges loss or fake edges under such circumstance as the thresholds being too high or too low in original Canny algorithm, this thesis proposes a new algorithm for Canny edge extraction with adaptive thresholds based on Bacterial Foraging Optimization algorithm and enhanced Otsu algorithm (ACE-BFA). This method is able to adaptively get the double-threshold according to the images characteristics instead of a man-made set, thus improving the precision of edge detection.(2) In order to compensate the issue of edges loss produced by traditional edge detectors, an effective Edge Linking method is proposed based on Fuzzy theory and Ant Colony Optimization algorithm (EL-FACO). The method analyzes the end points of all the line segments from the edge image obtained by traditional detection approaches. According to the gradient information of each pixel within clique, it calculates the membership matrix based on fuzzy logic. At last, it determines the transition function through gray level variation, membership and pheromone of each pixel, thus reducing the blindness of ant colony optimization, as well as improving the accuracy of indexing edge points. (3) Extracting edge of gray image though PCNN. Considering the large amount of parameters and the need for man-made set in segmentation cycle, this thesis proposes a method to judge cycle times and choose optimal segmentation result automatically through the minimum distance criterion. The experiments in the thesis verify the effectiveness of such method.