Research Of Processing And Content-Retrieval Based On The Images Of Pigmented Skin Lesions

Due to sunlight, moisture loss, sebum secretion, increasing age, genetic and other factors, various skin diseases like acne and pigmentation are caused, what's worse, those skin diseases may imperil human life. The World Health Organization assesses that more than 65000 people die from malignant skin every year. The diagnosis of pigmented skin is mainly based on macroscopic observation of doctors. The approach of traditional diagnosis relies on doctors' experience and ability. Besides, with the rapid increase of outpatient, doctors need to observe a large numbers of cases and read a large amount of image data, as a result, the diagnosis workload of doctors is unbearable. As the computer and digital image processing technology is applied in the medical field,the skin image analysis system and skin symptoms of computer aided diagnosis will come true.In order to achieve the goal of computer-aided diagnosis for pigmented skin image quickly and efficiently, we need to improve existing algorithms based on the actual environment:First, the preprocessing, content-based image retrieval of dermatological images, like eliminate noise, enhancement, automatic hair removal, segmentation, color space, image feature, is researched in depth. In addition, the similar measure and effective evaluation method for image retrieval are also analyzed for image feature extraction, content understanding, image retrieval and recognition.Second, deep convolutional neural network for learning hash functions that preserve semantic between similarity skin images is applied. In our approach, deep convolutional neural network is incorporated into hash functions to jointly learn feature representations and map from them to hash codes, which avoids the limitation of semantic representation power of hand-crafted features.Third, an effective scheme based on Hash-AP algorithm is used to solve the ranking of results. The algorithm is used to excavate the intrinsic link of data. Clustering data based on a measure of similarity is a critical step in scientific data analysis. The Affinity Propagation Clustering views each data point as a node in a network and then recursively transmits real-valued messages along edges of the network until a good set of exemplars and corresponding clusters emerges.At last, the pigmented skin image retrieval system is built. Users upload the pigmented skin and then feedbacks of the retrieval system not only show users the most relevant image but also provide the causes, symptoms, treatment and complications reference information.