The Processing And Detection Of HFMD Image Based On Dual-threshold Segmentation

The hand foot and mouth disease(HFMD)is a kind of infectious disease caused by enterovirus,mostly spread among children under five years old.Kids gathering places such as nursery and kindergarten are very vulnerable to the disease due to its strong infection and transmission ability.Therefore,a timely and accurate diagnosis of HFMD is very important for the detecting and controlling of the disease.This thesis proposes an image-processing algorithm,i.e.,the dual-threshold segmentation based on the R-G-B component distance,to automatically detect the HFMD images.Firstly,some fundamentals of image filtering and enhancement are introduced and the median filtering and fuzzy enhancing methods are applied for the pre-processing of the HFMD images.Then,the basic morphological operations are introduced and a scheme with gray-value morphological processing for differential gray images and binary morphological processing for binary images is proposed to improve the spots segmentation accuracy in the HFMD images.Finally,the dual-threshold segmentation based on the R-G component distance algorithm is proposed to extract the red spots in the images,which results in a higher segmentation accuracy than the Otsu's method and the maximum entropy method.The effectiveness of the proposed HFMD detection algorithm is evaluated on an HFMD image database collected in hospitals.Three metrics,i.e.,the missing rate,the false alarm rate and the F-measure,are used for the evaluation.We first investigate the optimal conditions for the single-threshold and the dual-threshold segmentation.Then we compare the detection results for different methods,i.e.,the single-and dual-threshold methods with their respective optimal conditions,the Otsu's method and the maximum entropy method.Results show that the proposed dual-threshold segmentation method achieves best performance.Take the Fmeasure for an example,20.09% improvement over Otsu's method,10.47% improvement over the maximum entropy one and 5.95% improvement over the single-threshold one have been obtained.