Research On The Method Of Skin Cancer Detection And Recognition In Dermoscopy Images

The number of skin cancer patients has been increasing in recent years.Melanoma account for a high proportion of skin cancer and melanoma has a very high mortality rate.If melanoma can be diagnosed at an early stage,the survival rate of patients will increase.However,nevus and melanoma have very similar appearance and symptoms,and it is difficult for doctors to diagnose melanoma by observation.In response to this problem,this thesis aims to design an auxiliary diagnostic method that can detect and identify melanoma,moles,and basal cell carcinoma in dermoscopy images,and achieve higher accuracy in the skin cancer identification stage.The main work of this thesis is as follows:1.The dermoscopy magnifies the skin cancer area and enlarges some factors that affect the observation of the dermatologist,such as hair and blood vessels.And the dermoscopy image is also easy to generate some noise in the acquisition.In response to these problems,the hair removal algorithm,Gaussian filter and Wiener filter are used in the preprocessing to process the dermoscopy image.In the experimental results,we can see that the preprocessing method designed in this thesis can effectively reduce the impact of noise on the segmentation results and feature extraction of the skin lesion area,and can improve the accuracy of the classification results.2.Current computer-aided diagnosis systems cannot effectively diagnose melanoma in the early stages,so it is still a challenging task to find better features to identify skin cancer in the initial stage of melanoma.To solve this problem,this designs a feature combination,which contains texture features and color features,and improves the extracted features,and finally introduces a feature fusion algorithm to process the features.This thesis first selects a combination of color features and texture features from the common color features,boundary features and texture features by using sequential floating forward selection(SFFS)algorithm.Then combined with the symmetry of the lesion area,the symmetry information is added to the selected color features and texture features.Finally,the multi-dataset discriminant correlation analysis(MDCA)is applied to fuse the selected features,and it is found that the fusion feature vector improve the performance of the computer-aided system.