Research On Image Segmentation Algorithm Based On Watershed Algorithm

In the field of image processing,image segmentation is a crucial technique for image follow-up processing.Image segmentation can divide the area of interest in the image,which is the basic step of image analysis,extracting image features and image understanding.The application of mathematical morphology in the method of image segmentation is one of the hot spots of image segmentation in recent years,and the typical watershed segmentation algorithm based on morphology is typical.The watershed algorithm is widely used in many fields because of its advantages,such as better extraction of target contour and simple operation.However,it also has the shortcoming of over-segmentation,which makes the segmentation result not ideal.The main reason is that watershed algorithm has a strong response to the noise and other interference factors in the image,so it is easy to produce over-segmentation phenomenon when the image is directly carried out watershed algorithm,resulting in the region of interest can not be well extracted.Although the researchers put forward the solutions of watershed segmentation pre-processing and watershed segmentation post-processing,the problem of over segmentation still exists.In this paper,the over segmentation problem caused by the existing watershed algorithm is studied.The main research contents are as follows:(1)Solving the problem of over-segmentation caused by the watershed algorithm is the premise of using the watershed algorithm for segmentation.In order to solve the problem of over-segmentation,this paper proposes an algorithm that combines the SLIC algorithm and the watershed algorithm.Firstly,the complexity of the input image is calculated,and the number of pre-segmentation super-pixels is calculated based on the size of the image.Then,SLIC is used to pre-processing the original image into super-pixel segmentation,which reduces the data redundancy in subsequent processing.Then,in order to eliminate the noise in the image and obtain more complete contour information,a method of adaptive threshold acquisition is proposed,and threshold processing is carried out on the gradient image of the pre-processed image.Finally,the image extracted by minimum mark is reduced by watershed segmentation algorithm to obtain better segmentation results.The experimental results show that the algorithm can effectively solve the over-segmentation problem and obtain a better segmentation result.(2)In order to better solve the phenomenon of over-segmentation,the deep learning and watershed algorithm are combined,but the existing algorithm is not well segmented on the edge of the image.In order to solve this problem,an instance segmentation method based on the improved depth watershed algorithm is proposed,first of all,the original image is converted to RGB,HSV and Lab color space image,and the image of three color spaces is obtained,and secondly,based on The idea of Color Net,three color space imagesare trained using Deep Lab V3 plus network,the results of three semantic segmentation results are obtained,and the split image is pixel-level fusion,and the image with rich pixel information is obtained to improve the image semantic division of the image.Finally,the resulting result graph and RGB original image are used as inputs of the watershed algorithm network based on deep learning,so as to improve the accuracy of image segmentation.(3)In order to better in the instance of the segmentation algorithm based on improved the depth of the watershed algorithm into a richer context information,puts forward the improved Deep Lab V3+ network,will join the rest of the network location attention module,better consider each location in the image of contact and relevance,reduce false segmentation,obtain better segmentation result instance.Firstly,the transformed image was trained with the improved Deep Lab V3+ network,and the semantic segmentation results were fused.Then the fusion image and the RGB original image are used as the input of the direction network and watershed transformation network to better utilize the semantic details of the image to obtain improved instance segmentation.Finally,the semantic segmentation results and the instance segmentation results are obtained by weighted prediction of panoptic segmentation.