Image Semantic Classification Method Based On Line Identification And Structure Module Identification

With the popularity of computer and smart mobile, people use a massive amount of images in their daily lives. Because the computer can not directly obtain image semantics from the image, there exists a semantic gap between the semantic and the image features. Computer can not understand the content of image from the two-dimensional image data.In this paper, we do image semantic classification by considering the relationship between points and lines.First,this article briefly describes the background of image semantic,the image feature extraction and semantic recognition algorithms. This paper briefly introduces the description method commonly used in image semantic features, including the global and local features. And then introduce canny edge detection algorithm and syntactic pattern recognition.Edge detection algorithm is used for image pre-processing to simplify the data. Syntactic pattern recognition is different from the statistical pattern recognition methods.Most of the current semantic image recognition algorithms are based on statistical pattern recognition.Syntactic pattern recognition method give us another way to think about the problem.The first image semantic recognition algorithm:a method based on the identification of the lines. Line recognition method are based on recognition the contour lines of the image. Algorithm steps:first use Canny algorithm to convert images to contour lines, and then propose a method to identify different lines from the image and distribute lines based on line length and the rate of angle change, finally make semantic classification between different images. The experiment proved the effectiveness of the proposed algorithm, experiments in the case requires only a small sample, it gained high recognition rate.The second algorithm, a method based on the module structure. It is suitable for image which have a clear structure and limited elements, such as text images. First Ⅰ define the element module, then find and extract the different modules, finally calculate the similarity with sample image. Both methods are focused on lines in image. The first method is applicable to all images, but the results have no meaning,it calculate the similarity with statistical distribution. The second method is only applicable to image witch have a clear structure, and it get result which have clear meaning. Finally, the paper introduces suitable environment of the two methods and their features, and what needs improving in the future.