Study On The Detection And Recognition Of Seawater Pearl Based On The Integration Of Information Entropy And Feature Extraction

Seawater pearl is a kind of important aquatic products in our country. It’s irreplaceable in the jewelry industry because of its shining and beautiful appearance. At present, the identification of seawater pearl is artificial in China, which largely relies on the experience of people so it’s easy to make misjudgment. Therefore, it’s significant to find a way to identify the seawater pearl automatically by using the technology of digital image processing and pattern recognition.Seawater pearls’physiological characteristics, such as colors and textures, are in the nature of regularity but also randomness and discreteness. Thus, this paper proposes a method which takes the information entropy, color and texture features into consideration to analyze and study seawater pearl image comprehensively. In the aspect of color feature, a method called multi-resolution block color histogram is proposed to resolve the problem of information redundancy which is led by the reason that traditional partitioned histogram didn’t distribute weight to each sub-block. Experiments show that this method not only retains the rotating invariance of block color histogram and color spatial information, but also eliminates the color information’s redundancy in some extent, and increases accuracy of its similarity being calculated. In terms of texture feature, the gray level co-occurrence matrix method is researched. According to seawater pearl’s texture characteristics, the most suitable structure parameters of seawater pearl gray co-occurrence matrix are acquired by experiments, and then the corresponding texture parameters are calculated. As to the multiple feature fusion, distribute the weights of colors similarity and textures similarity according to their own information entropy.Based on the above algorithms, extract the image feature of seawater and freshwater pearl and establish the feature database to do research. The sample is divided into three categories by K-means clustering:freshwater pearl, beige and white seawater pearl. In the process of recognition, two layers which is composed of textures filter and primary colors filter are designed. The purpose of textures filtering is to determine whether the image continues to participate in the detection:if the numbers of texture parameters are not in the threshold range, it can be convinced that this is not a seawater pearl image; if they are in the threshold range, the detection will be continued, select the feature library through primary color filter. Finally, calculate the characteristics similarity between the detected image and the images in the sample library. Then judge its category by the maximum similarity principle. Thereby, the category of the samples can be selected. Experiments show that the texture and color features fusion technology can improve the efficiency of image recognition.