Insect Image Segmentation Technology Based On Multiple Linear Regression

Image Segmentation is an important research in image processing and computer vision, and itis also the foundation of image analysis. The research of insect image segmentation algorithm is ofgreat significance to pest prevention and effective utilization of insect image information in thefield of biology. Therefore, insect image segmentation is widely concerned by domestic andforeign scholars in recent years.Multiple linear regression is a model of regression analysis based on mathematical statistics, itmakes use of several variables fitting to consist linear relationship to realize information processing,and is widely used in economic analysis and traffic forecasting. This paper analyses the limitationof image segmentation methods are applied to insect image segmentation, and is focused on theimage segmentation algorithm based on multiple linear regression model, according to theshortcomings of segmentation for many different kinds of color insect images, an improvedmethod is described. The major work and innovations in paper are as follows:(1) Building the multiple linear regression model of RGB three color boards to segment theinsect image. The determination of the parameter of the multiple linear regression model is the key,with the RGB three color boards sample information of insect image background to model canmake the insect target and background separation effectively. The method is applied to insectimages of which background have few colors, and achieves a more ideal effect.(2) Using the cosine norm to optimize the multiple linear regression model. Thedetermination of the parameter of the multiple linear regression algorithm based on RGB threecolor boards depends on the sample information of the image pixel blocks, the more the sampleinformation, the better the segmentation of image’s target area, but if the pixel information valuerange is large, it could cause a big deviation to the linear fit of the regression model. To this, thispaper optimizes the RGB three color boards information by using the cosine norm. Experimentsshow that the optimized algorithm is effective to linear fit of three color boards of sample pixelblocks, and it is suitable for the insect image of which background is more complex.(3) For segmentation of some insect images with shadow is not ideal, the optimized algorithmbased on multiple linear regression and transition region is proposed. The new method adopts thetransition region to realize the twice optimization for insect image based on the multiple linear regression, it solves the problem of boundary segmentation for some insect images with shadow,and improves the accuracy of insect image segmentation.