Sketch-based Image Retrieval Using Edge Histogram Descriptor

In today's Internet era, each detail of people's lifestyles such as communication and works, becomes a vast amounts of multimedia information. Image is an important information of media transmission and also the basis of video. Image cannot be replaced because it has become the most necessary part of our lives. How to search the correct image you want through the most convenient way becomes a key point that everyone pays attention to. Image retrieval is the main direction of multimedia retrieval nowadays, which means getting the semantics and similar images of the database by inputting some keywords, images,sketches, etc..Since twenty-first century began, the smart devices have become more and more popular, and they have also totally changed people's traditional lifestyles. In addition, the touch panel has become the first choice of input. Through touching the screen and tablet with fingers, people can input the sketches they want. Sketches are drawn by people so that they are not effected by light and have no color differences. Sketches include the information of the external shapes that people depict. The process is the same as human's perception to the shapes of the images. Therefore, it really draws a great attention via drawing some sketches on the Internet or finding similar retrievals in the database that is related.The article discusses the background and significance of scientific research in the field of sketch retrieval, introducing the current situation of form sketch retrieval work. It including the details of sketch retrieval throughout the process, pretreatment and related features. Then, the paper also does some distinctions, and analyzes a variety of edge algorithms, including classical and newly improved. These are about the ERH, SIFT, HOG and other characteristics of the basic characteristics. Moreover, it describes some methods to matching and retrieval algorithm of the overall evaluation.The paper is about applying Edge Histogram Descriptor to the sketch-based image retrieval(SBIR), proposing two ways of improving the feature extraction. The first one is that according to the multiscale edge detection theory, it is to be proposed in this paper that a new approach of Sketch Based Image Retrieval through using Nonsampled Contourlet(NSCT) transform and Edge Histogram Descriptor. By using NSCT transform decomposition of image to do the edge detection for all high and low frequency subbands, it extracts the edge histogram of the subbands that is used as describing the features of images and sketches. The second one is that according to the gradient field theory to improve Edge Histogram Descriptor. Through transforming to get images and sketches of gradient field, and it will generate 1000 random sample points in the gradient field. Then, it can extract Edge Histogram Descriptor from local windows that taking a random point as a center and drawing out a square. That is used as images and sketches that are describing the features.