Research On Segmentation Method Based On Image Histogram Construction And Threshold Determination

Image segmentation technology has wide and important application value in military application, medical image processing and biometrics, traffic safety, modern industrial automation and other fields. It is a very important research topic in the field of image processing and computer technology. The segmentation quality is the important foundation of effectively implementing image analysis and understanding. In the existing image segmentation methods, the threshold segmentation method is widely used and well known in research. The general idea of the method is that image information is used to build the histogram, and then a certain threshold method is used to get the best segmentation threshold, which is characterized by a simple implementation and calculation, stable performance and effective segmentation.In this thesis, the background and significance of the subject was introduced firstly, definition and classification of technology of image segmentation was synoptically introduced, and the wide of entropy threshold and Otsu threshold methods was introduced. Then the histogram of the building and the determination of the threshold were introduced, including seven building histogram methods and some threshold algorithms based on building the histogram, then using entropy threshold method and Otsu threshold method to find the best threshold. In this paper, paying more attention to building the image histogram, based on reviewing and summarizing existing scientific and technological achievements, some threshold segmentation methods were studied. Mainly includes the following aspects:One-dimensional Otsu and two-dimensional Otsu methods are classical algorithms, but dealing with an image whose histogram is heavy tail distribution or skewed distribution and threshold results may appear deviation. Segmentation effect is not very good due to the image histogram without some significant features when the segmentation target and background blurry; A method for image segmentation problem based on gray-level spatial correlation maximum between-cluster variance is proposed. In this method, the histogram of comprehensive utilization of pixels in the image gray scale and spatial distribution of information is structured, which the histogram can be more organizly distributed in order to use the Otsu threshold method to find the best threshold.Due to considering the gray level spatial distribution information, some image segmentation technologies based on entropy threshold can enhance the thresholding segmentation performance. However, they still cannot distinguish image edges and noise well. Even though GLGM(gray-level & gradient-magnitude) entropy can effectively solve the problem, it cannot segment effectively multi-objective and a complex image. So, image segmentation with multi-threshold of GLGM entropy based on genetic algorithm is proposed in this thesis. In this algorithm, integral figure method is introduced to reduce the search space, and the single threshold segmentation of GLGM entropy was further extended to multi-threshold segmentation. Lastly, the Real-code-GA method is used to determine the best thresholds.Local information of image can effectively describe the characteristics of the images, and it also has very important clues for image segmentation. So, an image segmentation method based on Gabor histogram entropy is also proposed in this thesis. In this algorithm, the histogram of containing scale information and direction information of the image was built, which we called Gabor histogram, and then we use the threshold entropy method to find the optimal threshold value; In addition, the idea of integral figure was used to reduce the dimension of looking for threshold;Finally, the three threshold method presented in this thesis are summarized and discussed.