Research On Image Similarity And Object Contour Location Methods

Image recognition is one part of the artificial intelligence and produces a large number of branches as the current academic front. This article selects the basis algorithms of image recognition to study and proposes new calculation methods for image similarity, edge detection and object contour location. The research content of this paper is as follows:(1) Image similarity is the basis of image retrieval. Retrieve specific image from the massive images not only need high accuracy, but also a very fast computational speed. For this, this paper presents a new approach for calculating the distance between two images in HSV color space. First, divide image into a limited number of blocks and get the dominant color histogram of each image block. Then, use the quadratic form distance function, which can make full use of the color similarity of the HSV color space, to gain the distance of two color histograms come from different image. In this step, a new method about the color similarity calculation is proposed. Finally, through a best-matched method, work out the total distance by weighting the distance value of each image block pair. The efficiency and performance of the approach are demonstrated on both simulated data and real image data sets.(2) Edge detection is the key of image recognition, therefore, the emergence of a number of classical algorithm for us to choose, such as Sobel operator, Roberts operator, Prewitt operator, Kirsch operator, Canny operator, SUSAN operator, etc. This paper puts forward an algorithm of edge de-noising and connectivity enhancing--Worm algorithm. First, use Canny algorithm to get the edge image. Then, Worm algorithm removes the clutter edge lines with maintaining the connectivity of edges. The experimental results show that the algorithm has a good performance in making the edge image clear and concise. Compared to the edge thinning algorithms, it can remove the clutter edge lines effectively and simply. To a certain extent, it also eliminates the influence of salt-and-pepper noise in an edge image.(3) The object contour location can be attributed to the image segmentation algorithm, used to obtain the edge of object. In order to achieve real-time requirements, by consolidating various algorithms, this paper presents a fast method to locate the object contour. First, convert an image from RGB to YCrCb color space and set a candidate point on the target object. Second, move the masks of SUSAN edge detector straight toward8-directions to detect object’s edge points. Meanwhile, the box-based codebook model will generate a color model for this object. Once the masks reach edge points, they will detect object’s contour along its edge by considering the information of SUSAN mask and object’s codebook model till finish the segmentation. For performance evaluation, this paper applied proposed method to a real-time hand gesture recognition system.