Illumination Processing Based On Phong Model And Its Application In Defects Detection For Billet

Defect segmentation is related to the image gray scale,and the uneven effect of the gray level affects the segmentation result,which results in the error of the rating.Therefore,it is necessary to deal with the light of the billet,improve the quality of the image,facilitate the extraction of defects,and prepare for the subsequent rating.This thesis is mainly about the studies of billet image to the illumination processing problem and the automatic evaluation of the defects such as center segregation.Firstly,based on the classification of the non-uniform illumination image,this thesis focused on the highlight characteristics billet image segmentation problem of incomplete defects flaw feature and analysis of the distribution of illumination;Using the classical illumination processing algorithm to processing the uneven illumination image,and analysis the result from both subjective and objective aspects,summarize the illumination processing method of each type of the image.Secondly,this thesis presents the background fitting method based on Phong model to handle the challenge of the illumination processing method of the highlight characteristics billet image.This method put forward the specific background illumination fitting scheme by generate the geometry model of the physical reflection of the lights,using the background subtraction to generate the illumination processed billet image,the flawed target is easy to be separated out by method of threshold segmentation,target segmentation is complete,so it could decrease the influence of illumination to the image,and analysis the illumination processed billet image from both subjective and objective aspects.Finally,in order to evaluating the flaws such as center segregation,the method based on BP neural network is put forwarded.This method,according to the description of the segregation defects evaluating standard,use the eigenvector based on the defect characteristic vector of the center envelope as the input vector of neural network,and setting up the appropriate network parameters,training the sample andevaluating the segregation according to the defective evaluation level to design the binary code of the network output,the experiment shows that this method can evaluating the segregation automatically and the result is accurate and reliable.