Study On Skin Lesion Segmentation Of Dermoscopy Images Based On Pigment Separation

Skin lesion segmentation,namely detecting skin lesions from the surrounding normal skin,is the core step of computer-aided diagnosis system based on dermoscopy images.In recent years,the skin lesion segmentation algorithm based on skin pigment concentration distribution has received widespread attention from scholars due to its low complexity and no need for learning and preprocessing.However,existing algorithms can only detect the simple skin lesions caused by the independent melanin or hemoglobin concentration disorder,such as freckles,erythema,etc.,and cannot complete the skin lesion segmentation on dermoscopy images with complex structures.At the same time,the existing pigment separation algorithm is highly susceptible to the light source,making the estimation results of the pigment concentration lack of accuracy and robustness,and ultimately affecting the performance of the segmentation algorithm.Therefore,which greatly limits the application of the skin lesion segmentation algorithm in clinical medicine.This paper conducts analysis and researches on the skin lesion segmentation of dermoscopy images based on pigment separation,the main contribution can be outlined as follows:(1)We propose a centering image-log-channel-difference algorithm based on optical density space(CILCD-ODS),which can effectively estimate the pigment concentration of skin image.To begin with,a relation model between the skin image and the pigment concentration is established,and the model is optimized to eliminate the interference of the imaging light source.Then,an independent component analysis method is applied to calculate the pure color concentration matrix.At the same time,of which the rationality is judged by the simpler and more efficient rationality verification method we propose.Finally,on the basis,we complete not only the estimation of the pigment concentration by using the reasonable pure color concentration matrix,but also the evaluation of the correctness of the results by introducing the pigment concentration distribution map.Experimental results demonstrate that compared with other similar algorithms,the proposed algorithm isbetter,more efficient and more robust to the light source.(2)We propose a skin lesion segmentation algorithm of dermoscopy images based on pigment concentration distribution(SLSDI-PCD).Which can detect the accurate edge of the skin lesion without any preprocessing of the dermoscopy images,therefore achieving the skin lesion segmentation of the dermoscopy images.Firstly,pigment separation is performed on the dermoscopy images based on the CILCD-ODS algorithm to obtain the independent concentration distribution maps of melanin and hemoglobin.Then,by combining the Otsu algorithm with the threshold segmentation algorithm based on the mean and standard deviation of pigment concentration,the melanin and hemoglobin lesions corresponding to the independent pigment concentration distribution are detected.Next,merging the previous two lesions by means of union operation to obtain more complete skin lesion areas of the dermoscopy images.Finally,which are post-processed using island removal and hole filling to ultimately obtain the more accurate segmentation result of the skin lesion.For purpose of quantitatively evaluating the performance of the algorithm,we calculate the different evaluation indexes of segmentation results at the foundation of popular evaluation criterion.And the correctness and robustness of the algorithm are verified by experimental analysis.