Study On Stick Figure Evaluation Based On Shape Matching And System Implementation

Through learning and training of stick figures, help to improve children's ability to identify the objects, improve the children's eye-hand-brain coordination ability, exercise the imitation ability and imagination of children. Parents usually use various kinds of man-machine interaction software for helping to learn stick figures, but these softwares only provide the function of simulation study, do not automatically evaluate the stick figures. Correct evaluation of the stick figures is the precondition of guided learning, in order to solve the problem of automatic evaluate the stick figures, this paper studied the algorithm of shape matching with no noise and noise, apply the shape matching technology to the stick-figure matching, and explore how to apply stick-figure matching to evaluate the similarity between stick figures, and then evaluate the stick figures. We say no noise, it refers to matching stick figures drawn by users with stick figures without noise saved in the system, conversely, the noise refers to matching stick figures drawn by users with stick figures extracted from the image with noise.To the stick figure matching without noise, extract feature points from the point sets of the shapes of stick figures according to the number of points set of the template, minimum the distance of centroid points between two shapes, then as the feature points as the origin of coordinates, calculate the shape contexts. Calculate matching cost between any two points of two stick figures, to find the minimum matching cost, to determine the optimal matching relations. compute the matching distance between the two shapes according to the optimal matching relations, then estimate the shape similarity. Experimental verification, the matching accuracy can reach above 80% used the method in this paper.To the stick figure matching with noise, extract the shapes of the stick-figure from images with noise, in this paper, based on the canny edge detection operator to detect the shape of the stick figure in image, and get the adaptive threshold use the Otsu algorithm, finally using 4 neighborhood contour tracking and the DP algorithm to extract the contour point set of the stick-figure, and the distance tolerance of DP algorithm is 1. To the stick-figure matching with noise in this paper, the author improved the shape context, and no longer computing the centroid point, change to calculate directly the logarithmic coordinates of the edge points, and use the edge points as the center of log-polar to calculate the histogram. Calculate the matching cost between either two points, set a matching cost threshold, they are similar when the matching cost between two points smaller than the matching cost threshold. The similarity is determined by the average matching cost of similar points and the percent of the similar points in the contour points. Experiments showed that the result of the matching is good when the threshold is set to 0.3.Based on shape matching and stick figure matching, a stick-figure learning system with automatically evaluate was developed in this paper. Score according to the matching results between the stick figures, and compare with people's score, the result shows that, the evaluation accuracy of this system reaches more than 78% and 67% in both cases.Finally, the author summarized the main research work in this paper and proposed further work plan according to the deficiency.