Research On Thresholding Algorithm Based Relative Information Of Valley Point

As a prerequisite for the analysis and processing of many image processing technologies,image segmentation is a very classic hotspot in the field of image processing. Due to theeffect of many factors, such as the imaging principle of the image, lighting conditions,imaging equipment, background environmental, extracting the interesting object from thebackground is rather difficult. So the image segmentation is still a research difficulty. Thethresholding technology has a very wide range of applications in the field of industrial qualityinspection, medical diagnosis, a military target detection, text and image processing, becauseof its simple principle of implementation, good segmentation results and high real-timeperformance. And many research workers propose a lot of image thresholding algorithm fordifferent situations in kinds of fields. Among these methods, the Otsu algorithm is favored bymany scholars, as its strong theoretical basis and the satisfying segmentation results. Andthere emerges a large number of excellent improved algorithm based on the Otsu method.This paper makes the appropriate analysis and improvement of the Otsu method in threeaspects, based on the in-depth study of the improved Otsu algorithm. The three aspects arerespectively the image pre-processing, single threshold segmentation, and multi-thresholdsegmentation. And the specific improvements are as follows:(1) As the actual image histogram contains the sawtooth and burrs information, thethreshold positioning and the efficiency of the algorithm are influenced. In this paper, a largenumber of image histogram are carried out a detailed analysis of experiments andobservations. And a kind of definition of the sawtooth and burrs information in an imagehistogram is given. Then the histogram can be filtered according to this definition. In otherwards the sawtooth and burrs points and the gray values in their neighborhood will be adjustaccordingly, in order to make the histogram adjusted as smooth as possible, while retaininguseful information.(2) The existing improved Otsu algorithm only takes into account the impact of thevalley point in the histogram, and did not notice the definition of valley itself is a relativeconcept. The valley exists with the relative crest. Therefore, this paper considers the relative relationship of the valley and its adjacent crests in the Otsu algorithm. The ideal thresholdshould be the very obvious characteristic valley gray value. In other words, the ideal thresholdshould have a large difference with its adjacent crests. So the threshold values obtained canavoid being affected only by the frequency of the gray value, which makes the threshold tendto those gray values only with small frequency but do not have the valley characters, resultingsegmentation fails. Finally the paper executes a large number of experiments in a sample setof the Berkeley image library, and compares the segmentation results with the improved Otsualgorithm, verifying the segmentation results and performance of the proposed methodsufficiently.(3) In order to ensure the threshold positioning accuracy while improving the efficiencyof the algorithm, this paper presents a fast recursive multi-threshold segmentation algorithm.First, to segment the image in a recursive way based on the algorithm previously proposed, aninitial set of the threshold values can be obtained. Then, screen the set of threshold valuesaccording to a certain rules proposed, to optimize the number of thresholds and thesegmentation results. Finally, a large number of experiments are executed with the brain MRimages in the sample collection of the Harvard University medical image library. And ananalysis of the segmentation results is made from the qualitative and quantitative aspects, toensure the applicability and effectiveness of the algorithm.