Shape Modeling And Image Segmentation Method Based On Convolutional Deep Learning Model

The image is the basis of the human being's cognition of the whole world,the objective reflection of the natural scene and things,and the collection of recording the shape of things,which itself contains a large amount of geometric information.Image segmentation,as an important way to obtain real-world information,is also an important topic in the fields of computer vision,pattern recognition,and image understanding.The ordinary image segmentation method divides an image into several regions by utilizing lower-level features of the image,such as color,grayscale,shape,texture,etc.,and the content in the same region has similar feature.It is also an important indicator to measure the effect of image segmentation whether it can completely and accurately extract the unique regions of the segmented image or the shape regions of human interest in different research.Due to the complex environment of the real world and the advantages and disadvantages of imaging technology,the segmentation effect is often affected by various types of noise,the shape of the segmentation is obscured by objects,and the target and background features are similar.At present,deep learning,as a complex machine learning algorithm,cause its multi-level neural network,can combine lower-level features of images to form more high-level attribute categories or features that are invisible to the human eye.Therefore,it is possible to extract the target shape information and more complex advanced information in the image.This paper proposes a method based on deep learning model to model shape based on deep learning model and image segmentation theory.For how to use shape information in image segmentation,a deep learning image segmentation model combining with traditional classical image segmentation framework is proposed to combine image low-layer and high-level information to assist image segmentation to improve the segmentation effect to obtain a complete and accurate image target shape.The research content is broadly summarized as follows:?1?For characterizing shape problems,a Convolution Deep Belief Network shape model is proposed for shape modeling.Convolution Deep Belief Network has evolved from Convolution Restricted Boltzmann Machine to extract more complex information and has the advantages of individual component models.Since the Convolution Deep Belief Network is a probability generation model in deep learning,it can model and characterize the joint distribution.The model structure,mathematical representation,construction mechanism,training and shape generation method are given in the paper.Through the CUB₂00₂011 and Weizmann Horse dataset experiments,the results show that the model is better and more effective than other similar deep learning models for modeling the shape.Euclidean distance is used to measure the difference between the generated shape and the original shape.?2?The traditional classical Condition Random Field image segmentation model with low-level image features such as image color combining with Convolutional deep learning shape model,a Convolutional Deep Boltzmann Machine image segmentation model is proposed.The target is segmented by combining high-level and detailed local features based on the image segmentation model of theCondition Random Field.The advantage of this model is that it can acquire the two-dimensional spatial structure features and detailed image local features of the input image,and the model contains a pooling layer to accept small translation transformations,which can reduce complex calculations due to weight sharing.The experimental results show that the proposed model uses the convolution deep learning model to make the input in the image segmentation flexible and obtain higher-order shape feature information.Therefore,in the process of segmentation of real-world complex images such as noise and occlusion,image segmentation can get better results.