Research On Order Morphology Edge Detection And Watershed Image Segementation

Image edge detection and image segmentation is a key step from image processing to image analysis and the research of thse aspects has been enduring, in addition to reflecting that in image analysis projects it occupys an important location, a certain extent, it shows that the work of the aspects considerable difficulty and challenging. At present, despite the research of the image edge detection and image segmentation has achieved many results, but there are still many problems to be resolved.The traditional image edge detection method has mainly taken the way of enhancing the image high-frequency components and the edge detection operator commonly used, are more sensitive to noise, in the edge detection at the same time it is very difficult to remove the noise better, and sometimes can not detect the image edge. Since traditional edge detection algorithm on the shortcomings, the new methods are constantly emerging, but there are still many problems, especially the coordination problems of the edge detection accuracy and noise immunity. Mathematical morphology is a nonlinear filtering method, which is applied to the edge detection by the basic idea that using the structure of certain elements to operater the basic operation of the original image and then subtracting the original image. Compared with the traditional algorithms, morphological edge detection in the image has a unique advantage, which is based on a collection of computing and has a non-linear characteristics of the edge detection can reflect on both the image collection of features, good image edge detection, but also to meet the real-time requirements, but also solve the coordination problems of the edge detection accuracy and noise immunity. Therefore, mathematical morphology for edge detection, both effectively filters out noise, but also retains the original details of the image and has good edge detection.Image segmentation is the first step in the image analysis, and a very important fundamental question in the computer vision research. The segmentation result has a direct impact on the subsequent image analysis, understanding and the landscape restoration. Image segmentation is commonly used in pattern recognition and image understanding and image compression and encoding. In recent years, many researchers dedicated to image segmentation method of research, but until now do not exist a common approach, and there is no one of the objective criteria to judge the success of segmentation. The extensive use of image segmentation, prompts people to search for new theories and methods to improve the image segmentation results and to meet the various needs. In recent years, image segmentation based on watershed algorithm gradually being people’s attention, and has become the research hotspot. Watershed algorithm has the burden of light, the high precision of calculation, but the watershed transformation will produce much many regions, and so as that the merger process computation is enormous, which need to address the time-consuming problem, with combining some time-saving way to the image segmentation will be a trend.This paper based on mathematical morphology focuses on the basic theory of image denoising, image edge detection and image segmentation. On how to use morphology transform ideas to better improve the image pre-processing, image edge detection and image segmentation to carry out quality problems, the paper research is to achieve the following objectives:(1) image de-noising to achieve good noise removal, image detail maintain the integrity and the output image signal to noise ratio significantly increased, while the adaptability and intelligence of the filter enhanced. (2) The image edge detection to correctly solve the edge of the presence or absence of true and false, and directional positioning. (3) The goal of image segmentation process of the correct extraction of interest, segmentation effect is good, fast. The work is laying a good foundation for conducting the latter image analysis.The paper has studied the conventional method and morphological transformation method in three aspects of the image pre-processing, image edge detection and image segmentation. Based on the advantages of morphological transform in the image processing, it has proposed separately by corresponding improvements morphology transformation model which confirmed feasibility by corresponding experimental.First, in the image pre-processing, it has researched and analysied of the common airspace and frequency domain filtering methods, as well as some of the existing morphology filtering methods, for the shortcomings to improve, it is proposed a multi-structure, multi-scale morphology filtering method based on adaptive genetic algorithm, the main consideration of the filtering window size, type and orientation of structural elements, as well as structural elements of the optimization of selection, using genetic algorithm optimization of structural elements, taking into account the convergence of genetic algorithm itself by using the elite’s strategy to retai. In addition to consider the choice of genetic algorithm parameters using adaptive strategy for maintaining population diversity, while ensuring the convergence of the algorithm. At the same time, it combines with adaptive weighted morphological filter ideas to build the multi-scale multi-adaptive weighted morphological filter based on genetic optimization. Use the peak signal to noise ratio of the image as an evaluation criterion of the image de-noising effect, and verify the method feasibility through experiments. This will lay a good foundation for the following image processingSecond, in the image edge detection, according to the three indicators of evaluating the edge detection performance strengths and weaknesses, namely, SNR, positioning accuracy, and single-edge response, and combined with the sequence of morphological characteristics, this paper proposes the following improved algorithm that combining the complex order morphology transformation and anti-noise-based morphological edge detection, construct a new kind of gray-scale image edge detection method, establish the edge detection oprators, improve signal to noise ratio, by order morphology transformation of the percentile values to control the effect of edge detection the same time, introduce the structural elements of different orientation to match the edge of the direction of diversity, improve the traditional mathematical morphology edge detector, thereby improve positioning accuracy, accurate positioning at the same time to reach the edge of the purpose of better noise suppression. They constructed three kinds of edge detection operator can not only better suppress image noise, and can more accurately detect the edge of the image. Respectively, Salt and pepper noise, Gaussian noise and contains a mixture of the two images in noise, edge detection, experimental results have compared to the commonly used edge detection operator and the general order morphology transformation, which verifies the edge detection model feasible proposed by this paper to a certain extent.Finally, in the image segmentation, it summaries the commonly used algorithms of the image segmentation, including the threshold value method, the regional growth and division consolidation method, edge detection and watershed segmentation based on morphology. These methods have their shortcomings. It focuses on the watershed algorithm, which is a classic and effective way to the image segmentation with the fast, effective and accurate results which could be found to win people’s attention. However, it exists the excessive division of watershed algorithm, for this problem, this paper presents an effective method of image segmentation, integration watershed algorithm and dynamic particle clustering, the first to use multi-scale multi-structure morphological transform filtering based on the adaptive genetic algorithm to the image pre-processing, to a certain extent filtering noise, and then use the Vicent algorithm split, for the over-segmentation problem, proposed a connectivity constraint with the dynamics of particle clustering and at the end to make appropriate follow-up treatment. The segmentation effect under the method is compared with the results of using different filters watershed segmentation, using thresholding and region growing segmentation, and the watershed segmentation with markers, the region effect is compared with the direct watershed, the traditional watershed segmentation algorithm for region merging based on boundary strength, and the comparison shows it is necessary that the pre-segmentation, and the filtering method proposed in this paper to pre-process is effective at the same time constraints connected with the dynamics of particle swarm clustering of regional consolidation and algorithms from the effect on the execution time is relatively good, and also verified the initial split of using Vicent algorithm coupled with the dynamics of particles with a connected bound clustering in terms of region merging for image segmentation is valid.The research of appling morphological transformation to the image preprocessing, edge detection and image segmentation has important theoretical meaning and application value, in this paper research, it should be for further research and exploration in the self-adaptive of the structural elements in the morphological transformation, the algorithm versatility and running efficiency, etc.