Dermoscopy Image Segmentation Combine Saliency And Adaptive Threshold Based On Wavelet Transform

According to the World Health Organization,the morbidity and mortality of skin cancer are increasing year by year,which is a serious threat to human life and health.Melanoma,the most deadly kind of skin cancer,is the main cause of the continuous increase in the mortality rate of skin diseases.Fortunately,if the malignant melanoma make timely diagnosis and clinical intervention in the early stage,it can greatly improve the patient's cure rate.Using the dermoscopic technique to observe the living cortex,analyze the rich texture and pigment characteristics of the dermoscopy image,it can quantitatively evaluate the severity and development trend of the lesion area,which will help doctors to diagnose the disease efficiently.The precise skin lesion segmentation is not only the key technology for dermoscopy image analysis and processing,but also the premise and guarantee for subsequent skin disease analysis and diagnosis.In view of the difficult challenges such as low contrast and blurred boundary in the dermoscopy image,this paper launches the research on the automatic segmentation of the dermoscopy image.The main contributions are summarized as follows:1)In order to effectively segment the lesion area,this paper proposes an unsupervised skin lesion segmentation method,which combines saliency and adaptive wavelet threshold to obtain more accurate skin mirror image segmentation results.Firstly,the color saliency map and the brightness saliency map are merged to enlarge the difference between the lesion area and the healthy skin to obtain the lesion area of interest.And then combine the saliency map with the adaptive threshold based on the wavelet transform to further improve the segmentation effect of the lesion region.2)Considering that the application of the saliency model on the dermoscopy image does not take into account the subjective high-level information such as human vision.In this paper,we use eye tracking technology to acquire eye tracking data,and it collects the eye tracking data from fifteen observers.The eye tracking database of the dermoscopy image is constructed by these data and then it constructs a skin lesion saliency model based on eye movement information.Firstly,we use the eye tracking fixation point and the eye movement gaze duration to acquire the foreground template composed by the eye movement region,and then the eye tracking saliency map is constructed by selecting the foreground template and the background template.Finally,it combines the eye tracking saliency,color saliency and brightness saliency map,which fusion eye tracking information to improve the segmentation performance.Therefore,we propose adermoscopy image segmentation algorithm based on fusion saliency and adaptive wavelet threshold.3)We conduct a series of experiments on two public datasets PH2 and ISBI2016 to evaluate the performance of segmentation algorithm.The effectiveness of the proposed method is verified from different aspects through various evaluation indicators.The quantitative results and visual effects of the proposed method on two common datasets indicate that our method is superior to most state-of-the-art methods.