Investigation of the effect of noise reduction techniques on dermatoscopy images

Skin diseases are diseases that need to be correctly diagnosed and treated at an early stage. Computer-assisted automated methods are needed in the field of dermatology to provide early diagnosis and treatment. Studies conducted in this direction have generally focused on the analysis and classification of dermatoscope images or performing lesion segmentation from images, and the results obtained from these procedures have been examined and evaluated. However, it has not been emphasized how successful the noise reduction techniques applied as a pre-processing stage are. In addition, in the studies in the literature, different classification or lesion segmentation approaches have been used together with different noise reduction techniques in automatic methods developed with dermatoscope images. Therefore, it is not clear which noise reduction technique is more effective and the effect of these techniques on the results is unclear. In order to solve this problem, within the scope of this thesis, the performances of noise reduction techniques applied on dermatoscope images were examined and compared, and the most appropriate technique was determined. In addition, a method that has not been used before to reduce noise from dermatoscope images is also proposed in this thesis. The findings obtained with three different evaluation criteria in our studies showed that the noise reduction process with the proposed method can be performed with a higher performance than the methods applied in the literature. It will help dermatologists to get more successful results from automatic detection, segmentation or classification approaches of skin lesions to be developed using the proposed method and to make more accurate evaluations in the diagnosis and treatment processes of these diseases.