Research On Image Segmentation Algorithms Based On Renyi Entropy

In the process of industrial production, with the scientific progress, industrial production and product testing were more and more automated, in the process, machine vision was applied more and more extensively. The of technology image processing as the basis of machine vision, plays a vital role. The treatment effect of image processing directly affects the production quality and quality analysis testing. At present,? the digital image processing technology has been applied to all walks of life, such as industrial production,?engineering,? medical research, military research and so on.? Gradually, digital image processing has developed into a discipline.?In the digital image processing, the technology of image segmentation technology is an important component of digital image processing. Good treatment results can be got by using good segmentation algorithm, as can greatly improve the quality of detection capabilities, so the research of image segmentation is important and significant to study.Image segmentation technology as a key part of digital image processing, is defined that according to the characteristics of image itself, the image is divided into several regions each of which has different characteristics, while the target area which people are interested in is extracted, this technology and process is called image segmentation. On the basis of image segmentation, the follow-up work such as data statistics, feature extraction, parameter measurement and so on can be carried out. As a result, the effect of image segmentation directly impact on the quality of the entire digital image processing, so it is very necessary to research image segmentation algorithms and with the development of the scientific technology, images will become more and more complex so that the demands for the segmentation algorithms will get higher and higher, which requires researchers to study more efficient and better algorithms to keep better quality of segmentation.Today, the image segmentation technique has been widely used in the quality inspection of industrial products, whose role cannot be replaced. Whether in the agricultural, industrial, medical or military research, image segmentation techniques play their unique role. For example, in agriculture, the segmentation techniques are used in quality inspection of rice, quality inspection of grain appearance, fruit quality inspection and classification, in the industrial field, these techniques are used in the detection of copper defect, PCB quality, products surface defect and leather tanning, in addition, in the quality detection of viscose filament, detection of solar panels crack, the density detection of printing net dot and so on.Viscose filament yarn is the high-grade textile fibers, which is produced by cotton linter or wood fiber as the basic raw material. In the chemical fiber industry, the viscose filament possesses superior performance such as good hygroscopicity , softness , skin-proximity, etc. The market demand of the high quality viscose filament continues to rise. The image segmentation algorithms discussed in this article are intended for Viscose filament section images and to obtain better segmentation results, thus, which can help and promote the quality detection of viscose filament. Finally, it will come true that machine vision technology used for quality detection replace original manual method in chemical industry.The image segmentation algorithm discussed in this paper are mainly used for viscose filament section images and printing net dot images, designed to work out segmentation algorithms which can process these two types of images and obtain better segmentation results. The image segmentation algorithms based on Renyi entropy selected in this article has a good segmentation performance. The destination of this paper is to study and design the key algorithms for viscose filament section images and printing net dot images on the basis of existing image segmentation algorithms through a large number of simulation experiments, then these kinds of algorithms are analyzed and compared in order to get the best segmentation results.The research on image segmentation algorithm began in the 50s last century, many researchers in the field achieved a number of research results through continuous efforts and overcoming all difficulties, so various algorithms have been proposed one after another. As science advances and various disciplines cross in depth, many theories in other areas were applied to image segmentation, such as fuzzy theory, wavelet transform, clustering, mathematical morphology, etc. These specific theories combining with the traditional segmentation method form new segmentation algorithms. Similarly, Renyi entropy discussed in this article is a kind of entropy in information theory, which combining with image segmentation method formed the segmentation techniques based on Renyi entropy. Information entropy theory was applied to image segmentation in the 80s of last century, in this paper, some improvements and innovation based on previous research were carried out, and better results has been achieved.This article focuses on the image segmentation algorithms based on maximum Renyi entropy. Firstly, in this paper, the classification of image segmentation and the basic ideas of image segmentation algorithm were described, including the algorithms based on threshold, based on the boundary, based on the region and based on the region, totally, there are four types of algorithms. Meantime, several common image segmentation algorithms based on threshold were described, such as bimodal histogram method (also called mode method), iterative threshold method, the minimum error segmentation method, Otsu method, etc., and the thinking and math basic of every algorithms were given. After common algorithms were introduced, the relevant knowledge of entropy in information theory was introduced, including the definition of entropy, the origins and definitions of Renyi entropy. Information entropy applied to image segmentation began in 80s of the last century. In this paper, the basic principles of the image segmentation algorithms based on maximum entropy of were introduced, and the image segmentation algorithms which are often used based on information entropy and their development were described.In this paper, firstly, one-dimensional image segmentation algorithm based on maximum Renyi entropy was briefly introduced, including algorithm ideas and implementations. On this basis, the image segmentation algorithm based on two-dimensional Renyi entropy was described in detail. Two-dimensional histogram is defined as being composed by gray value of image itself and neighborhood average gray value. In the Two-dimensional histogram, each point represents the probability of the appearing vector (gray value of image itself, neighborhood average gray value), then on this basis and combined with Renyi entropy, two-dimensional image segmentation algorithm of maximum Renyi entropy can be got. The fast recursive formula and its basic implementation steps of this algorithm were described after the basic idea of two-dimensional segmentation algorithm based on Maximum Renyi entropy was discussed. Based on the above discussion, in this paper, I have made some improvements on the two-dimensional maximum Renyi entropy algorithm on the basic principles. When the original two-dimensional algorithm is used to segment an image, a segmentation threshold value can be got from the whole image. It is easy to produce error segmentation in some regions of this image by using this segmentation threshold value, the original algorithm has been improved in the article in order to reduce the false segmentation rate, After improved, an image can be divided into a number of regions based on the specific circumstances of the image when an image was segmented, and there is its own segmentation threshold in each region, so the correct detection rate is improved because there are several segmentation threshold values in an image.After the two-dimensional image segmentation algorithm based on maximum Renyi entropy was discussed, the image segmentation algorithm based on three-dimensional histogram and maximum Renyi entropy was presented in this paper. In three-dimensional maximum Renyi entropy algorithm, neighborhood median gray value is added into the histogram on the basis of two-dimensional histogram, that is, the two-dimensional histogram was expanded into three-dimensional histogram and the neighborhood median gray value as the third dimension of the three-dimensional histogram. So, the three-dimensional vector is (gray, neighborhood average gray level, neighborhood median gray). As the addition space factor considered, better segmentation results can be obtained. In this paper, in the definition of three-dimensional histogram, the calculation method of neighborhood median gray value was improved, the weighted neighborhood median replaces the original standard median gray, the improvement of this algorithm is able to keep more edge information. Three-dimensional algorithm also greatly increases the computing time of the algorithm because of considering more space factors. To reduce the algorithm computation time, the fast recursive formula and its concrete implementation steps of three-dimensional algorithm were presented. Recursive algorithm is that three-dimensional vector is converted to sum of two-dimensional vectors, so it is necessary to increase storage, the cost of reducing time is to use smaller space.A large number of experiments have been done,?viscose filament fiber section images and printing net dot images as the experimental pictures.?Experimental results show that image segmentation algorithm based on two-dimensional?maximum Renyi entropy?is better than the classical Otsu algorithm for the above pictures,? and also better than the two-dimensional Otsu method. The handling effect of two-dimensional Renyi entropy algorithm is better than effect of one-dimensional entropy algorithm. The? improved two-dimensional image segmentation algorithm presented in this paper is better than original algorithm. Of course, considered more space factors, the longer computation time than one-dimensional algorithm is necessary. The segmentation algorithm based on three-dimensional histogram and maximum Renyi entropy presented in the paper, compared with the above mentioned algorithms, is best for the segmentation results. But its shortcoming is longer processing time. By using recursive algorithm, the processing time?is still within the acceptable range.