Research And Application Of Image Recognition Based On Three-way Decisions

With the development of computer network technology and the emergence of digital videos and images,the image recognition has become a hot issue in the current intelligent information processing.The traditional methods for image recognition were based on the low-level information and focused on detecting the key features by regional segmentation.Since the general method of image segmentation was regular rectangle-based,it was problematic to divide the same target into different sub-blocks.In addition,the general image recognition methodology usually ignored the fuzzy part of the image feature,resulting in incomplete recognition and low recognition performance.To address the above problems,this thesis focused on the following studies:1.Constructed an image recognition classifier based on the three-way decision mechanism.Due to the conventional rectangle-based segmentation methods,the same sub-target could be accidentally divided into different sub-modules.To address the drawback of mentioned methods above,a delayed decision with three-way decision mechanism and the corresponding classifier were proposed.Firstly,a delayed decision and re-determination process was applied to the fuzzy information;secondly,the decision was iteratively updated by joining the new conditions that was obtained from the classification results of the previous epoch/step;finally,through a limited iteration,the proposed method could classify the same target as close as possible to the same sub-module.Thus,the error produced by the conventional method was decreased,and the segmentation accuracy/performance was improved.2.Proposed a multi-objective SVM(Support Vector Machine)image recognition method based on three-way decision mechanism.Throughout the comprehensive literature review,the problem of classification caused by image segmentation in image recognition was found that the common SVM classifier only had binary(“positive” and “negative”)classes so that the setting of the image classification criterion could not satisfy all conditions.To solve the above problems,this thesis started to improve the classification algorithm,inspired by the human thinking mode,to combine the three-way decision with SVM classification.We designed an image recognition method based on three-way decision and multi-objective SVM,and its feasibility and the accuracy were validated by experimentation.3.Proposed a CNN(Convolutional Neural Network)image recognition method based on three-way decision mechanism.The problem of the traditional image recognition was the neglection of useful information.The proposed image recognition model deployed the three-way decision mechanism with the delay decision to make the fuzzy information gaining the fully classification processing.It was more efficient to use the available information,and accurate in segmentation and recognition.In order to make the process of image inferencing more consistent with the mode of human thinking,the concept of delay decision was added to the decision-making progress in image recognition when the image recognition model was built.The purpose of maximizing the useful information of the image was achieved by adding the decision condition by iteration until stable(convergence of classification).Based on the traditional CNN image recognition method,a CNN image recognition method based on three-way decision was designed.The feasibility and accuracy of the method were validated by experiments.