Research Of Colorful Image Segmentation Based On Neural Oscillatory Network

Previous studies have shown that neurons in the visual cortex of the cerebral cortex complete the function of reception,transmission and encoding of information.And it indicates that cognitive function is realized by the oscillation synchronization between brain neurons.A features are encoded by the synchronized activity of neurons in different regions of the cerebral cortex,in other words,a collection of synchronized neurons encodes a feature of the object.Therefore,the third generation artificial neural network,i.e.,neural oscillation network,is more in line with the physiological and functional structure compared with the first and second generation artificial neural networks in which the perceptron and activation function work as computing units.The structure has been widely and successfully applied in the field of image processing.Being a typical neural oscillation network,Kuramoto model plays an important role in complex network,clustering,image processing and other fields due to the convenient construction and "synchronization" capability.What is particularly striking is that the Kuramoto model groups the data points with a similar feature by the simple coupling relationship in the evolution process,and then obtains the number of data categories.However,this model could not directly output labels of data because its physical mechanism is based on phase oscillation,and the phase is continuously distributed in the range of [0,2?].Meanwhile,the process of phase oscillation does not consider the spatial relationship,it results that the model could not obtain the desired segmentation on complex images.Considering the above problems,we improved and expanded Kuramoto model to realize the segmentation of color images,and proposed two unsupervised segmentation methods for extracting multiple objects as follows:1)Based on clustering and the localized Kuramoto model,this paper proposes an image segmentation method.Since Kuramoto model has the computation cost when applied to image segmentation,and does not consider the spatial relationship between pixels,we persist the capability of phase synchronizing,localize the similarity matrix to reduce the computation cost;employ clustering method to introduce the relationship into segmentation,and realize automatic the classification of the continuous phase map.The proposed method improves the Kuramoto model by a localization paradigm,and realizes multi-phase segmentation of color images via combing clustering method.2)Based on an active contour model and Kuramoto model,this paper proposes an image segmentation method.For the lack of consideration of geometric features in Kuramoto model,we introduce length term and region smoothing term into the energy functional of Kuramoto model to constrain the production of the phase map;meanwhile,apply active contour model to maintaining the geometric features by curve evolution,and track the accurate boundaries of objects.By introducing geometric regularization terms into the energy function,we expanded Kuramoto model,employed active contour model to track the boundaries of objects,and finally realized the multi-phase segmentation of color images.In brief,our paper improves and expands Kuramoto model,combines it with clustering method and active contour model respectively to realize multi-phase segmentation of color images,and contributes to its application in image processing and analysis.