Research On Multilevel Image Segmentation Based On Swarm-Based Intelligence Algorithm

With the rapid development and application of computer,handheld terminal,network and multimedia technology,Multimedia data have found their way into people's daily life,and in particular the image and video data which have shown explosive growth.With such massive image data,image processing has become not only a research topic but also a basic issue which must be confronted with in the era of information.As an effective means of image processing,image segmentation has become a significant step and prerequisite for image analysis and understanding,which has found direct applications in many areas such as industrial automation,product testing,text recognition,face recognition,intelligent transportation,alien exploration,remote sensing satellite image processing,aeronautics and astronautics,biology and medical engineering,network and computer vision,sports and agriculture,etc.Among the image segmentation algorithms,mulitilevel image thresholding as one of the most direct and simplest methods has attracted increasing attention worldwide.As the computational complexity of these methods is relatively higher,the swarm-based intelligence algorithms are widely used to improve the efficiency of segmentation.Drawing on the advantages of artificial bee colony algorithm and the gray wolf optimizer in processing speed and stability,in combination with the fuzzy theory and fuzzy logic,this dissertation makes a study of the objective function selection of the multilevel image segmentation,the improvement of the swam-based intelligent algorithm and the fuzzy membership degree initialization and aggregation methods.Based on the correlation characteristics of multi-level image segmentation,this dissertation improves the running time and convergence of artificial bee colony algorithm,and enhances the stability of gray wolf optimizer.Considering the spatial position independence of multilevel image thresholding,the study also improves the image segmentation precision by fuzzy initialization and local fuzzy information aggregation to achieve the final purpose of the fuzzy multi-level image segmentation mechanism aiming at promoting the segmentation speed,ensuring the stability of the algorithm and improving the segmentation precision.The main contents and contributions of this dissertation are as follows:(1)Image thresholding based on modified artificial bee colony algorithm.As the number of thresholds increases,the complexity of the algorithm increases dramatically.Thus considerable research work has been done to deal with the problem through the swarm-based intelligence algorithm.On the basis of the analysis of the swarm-based intelligence algorithm,a multilevel image segmentation method based on improved artificial bee colony algorithm is proposed.And the running time and convergence of the artificial bee colony algorithm are optimized by improving the neighbour search strategy of the onlook bees.The contrast experiments show that the artificial colony algorithm enjoys high performance in image segmentation quality,running time and convergence.(2)Image thresholding based on modified discrete gray wolf optimizer.In response to the problem of multilevel image segmentation,the dissertation first proposes the discretization scheme of gray wolf optimizer,and then improves the wolves attack behavior through the weight mechanism to achieve the purpose of enhancing the current optimal solution in the iterative process of participation and importance.By analyzing and comparing the Kapur and Otsu objective functions,the experimental results show that the improved gray wolf optimizer achieves satisfactory performance in image segmentation quality,optimal objective function value and stability.(3)Image thresholding based on fuzzy theory and swarm based intelligence algorithm.With the test results of Part(1)and Part(2),firstly the Kapur's entropy is fuzzified by trapezoidal membership function,and then the two swarm-based intelligence algorithms are optimized to obtain a set of thresholds.Finally,the fuzzy segmentation of the image is realized.On the basis of this,the optimal thresholds are taken as the fuzzy centroids,which provide the basis for the neighborhood fuzzy aggregation of the pixels.And then the optimal thresholds and the optimal objective function of a variety of fuzzy segmentation methods are compared,which verify the efficiency of multilevel image segmentation based on swam-based intelligent algorithm and fuzzy theory.(4)Image thresholding based on swarm intelligence and local fuzzy information aggregation.On the basis of fuzzy swarm intelligence algorithm,through the fuzzy membership degree initialization and fuzzy membership degree aggregation,a fuzzy multilevel image segmentation scheme based on improved qucik artificial bee colony algorithm and improved discrete gray wolf optimizer is proposed.In the process of algorithm execution,the fuzzy values are assigned to each pixel by the pseudo trapezoidal membership function,and then fuzzy information are aggregated with median,mean and iterative mean respectively so that the fuzzy multilevel image segmentation is realized finally.A series of comparative experiments have proved that the median aggregation achieves the best segmentation effect and is superior to other fuzzy and non-fuzzy methods.The experimental results also show that the modified quick artificial bee colony algorithm is optimal in terms of time efficiency,and the modified discrete gray wolf optimizer is more prominent in terms of stability.