Hierarchical Feature Matching Of Fault Images In TFDS Based On Improved Markov Random Field And Exact Height Function

Trouble of moving freight car detection system(TFDS)is a dynamic fault image detection system based on machine vision developed by China.Because of the constant color and complex background of TFDS images,a hierarchical feature matching method is proposed based on improved Markov random field and exact height function by using spatial hierarchy and shape feature of fault images.In order to realize the automatic detection,the proposed method divides fault detection into two parts: hierarchical model building and shape matching.Considering the pixel space interaction,an image segmentation algorithm named fast adaptive Markov random field(FAMRF)is proposed based on MRF combined with the Pyramid model and affinity propagation theory.Firstly,both wavelet transform and MRF theory are applied to build a multi-scale hierarchical model.Then,histogram smoothing technology and affinity propagation are introduced to assign the number of layers and to achieve the adaptive image segmentation with modified K-means algorithm.Finally,the standard deviation differential of pixels is considered as iterative criterion to further enhance the speed and robustness based on energy stability in the process of segmentation.Experiments on both McGill image database and Weizmann image database demonstrate the effectiveness and high accuracy of the proposed method.The FAMRF is slightly superior to the original MRF,and the segmentation efficiency increased by over 40%.In view of the practicality of shape matching in image recognition,an exact height function(EHF)shape descriptor is proposed on the basis of height function(HF).Firstly,the contours are extracted from target shapes,and then EHF descriptors are constructed based on the precise height values for each sample point on contours.A smoothing technique is adapted to reduce the descriptor dimensionality.Then,the modified parallel dynamic programming is employed in matching stage.Finally,the shape complexity analysis is used to improve the matching accuracy.Based on point geometric feature saliency,the shape precision definition is introduced to further analyze the influence of local deformation and edge noise on shape description.The matching experiment has been conducted on the database of MPEG-7,Swedish Leaf,Tools,ETH-80 and noise experiment has been conducted Kimia99 dataset.Experimental results indicate that the proposed algorithm is highly efficient and its matching time is only 12.5% of HF shape descriptor.The highest retrieval ratio can reach 90.36%?95.07%?94.29% and 89.90% in above experiments respectively.Retrieval performance and robustness of EHF are better than HF and other important algorithms.Moreover,the anti-noise performance of EHF algorithm is superior to the original HF descriptor.According to the characteristics of TFDS image,based on hierarchical model and shape matching,hierarchical feature matching algorithm for fault images in TFDS are proposed to achieve automatic recognition of air brake system malfunction,bogie block key missing,the loss of high wear synthetic brake shoe and the absence of fastening bolts on brake beam.The proposed algorithm achieves high rate and good robustness,and it can be effectively applied into the fault detection for TFDS images.