| With the upgrade of hardware device and the advancement of image processing theory,solutions based on machine vision have been widely used in various fields,including object measurement,object recognition,vision-based mechanical control,product surface defect detection,etc.owing to the advantages of high efficiency,precision,and intelligence.As the bottom part of image processing,the precision of image segmentation greatly affects the accuracy and correctness of high-level image process.However,traditional segmentation algorithms are not ideal for image segmentation with blurred edges and weak contrast.In this paper,button is the research object.We use the image segmentation algorithm based on level set in order to improve the segmentation accuracy.We enhance robustness to environmental changes by applying guided filter and gamma gray transformation.Furthermore,downsampling,Canny operator and the gray projection are adopted in the task of rough segmentation,which effectively decreases amount of calculation.For the buttons with different optical characteristics,we study the level set methods respectively based on CV,local pre-fitting energy and local structure tensor,and determines the best model method for effective segmentation of different buttons.A regularization term is added to the CV model.In the LST model,the energy function is optimized so that the number of iterations is reduced,and the segmentation accuracy is slightly improved.We use 3 kinds of evaluation criteria.Among them,the DSC coefficient is coefficient is over 98%.Moreover,this paper analyzes the key parameters involved in the models and recommends the range of values to help better results.All of the performance including illumination adaptability,anti-noise and running speed of the algorithm can meet the requirements of industrial online segmentation.In this paper,the algorithm is implemented in C language,the processing speed of the 640×480 pixels image approaches 33 fps when tested on the PC platform. |
PDF Full Text Request