Skin Image Processing And Analysis Based On Pigment Separation

Researches on human skin analysis have been conducted in many fields, such as Computer Vision, Cosmetic and Dermatology. Compared with in-vivo skin research, digital skin image based methods are much safer and convenient, and are easier to analyze and store. Differing from research on general-purposed digital images, image based skin research focuses on the understanding of skin structures and optical properties. Complex phenomena can happen during the interactions between incident light and human skin. Incident light can be reflected on surface, or enter sub-surface and be scattered, reflected, refracted, absorbed and transmitted. Except for the reflected light on surface and the reemitted light from subsurface, the other incident light is absorbed by skin pigments located in skin layers, in which melanin in epidermis and hemoglobin in dermis play dominant roles and mostly decide the skin appearance. Various skin diseases can be induced due to lack of melanin and hemoglobin, pigment deposit and abnormal change of pigment concentration. Accordingly, most researches on human skin always put their emphasis on above two kinds of pigments.In this dissertation, we focus on pigment analysis based on skin images. Chapter 1 introduces existing researches on human skin. Chapter 2 summarizes the most basic knowledge - skin structure and corresponding optical properties- in skin research. As the most important chapter, Chapter 3 gives detailed introduction to the pigment separation technology with a single image as input. Considering that the existence of specular reflection on skin surface can produce local highlight on image and consequently leads to incorrect separation result, Chapter 4 presents a novel algorithm for pre-processing of pigment separation, and detecting and removing the highlight on skin image. In the two chapters follows, analysis after pigment separation and possible applications, as well as corresponding experience results are presented. The dissertation ends.with.a summary of the dissertation as well as the prospect of the relevant potential applications and future research directions.Contributions of the dissertation are listed as follows:1) Classifications of relevant studies on human skin and relationships between the classificationsThe object for human skin research may be in-vivo skin or non-invasive skin image, or even the real human body in specified environment. Different objects and application fields induce the diversity of human skin research. As far as we know, no research has been devoted to the classification of various research directions regarding human skin. In this dissertation, human skin researches around IT fields are classified into two categories and three dominant classes (See Fig. 1-3) that include many sub-classes. In addition, the dissertation gives the relationships between the sub-classes. The classification shows completely the background of skin research and highlights the most important points which should be primary considered in skin research. What's more, it also lies out a blueprint of possible research directions in the future.2) Detailed introduction to skin structure and optical properties Understanding the skin structure and optical properties are the primary purposedfor skin research. Skin structure is the predominating goal since optical properties are determined by skin structure. There are only few researches, mainly from Chinese scholars, on this subject. Chapter 3 of the dissertation introduces the three-layer skin structure, which is composed of epidermis, dermis and subcutis. Consequently, complex optical phenomena are induced when light interacts with skin layers. With the help of mathematical formulae developed, Chapter 3 details all kinds of interactions between incident lights and skin layers, such as reflection, transmission, refraction, diffusion lights and absorption, and summarizes the optical features in each layer.3) A novel algorithm for decomposing skin pigment based on single imageThe only algorithm existed in decomposing skin pigment from input image has many shortcomings. It's not robust and can't get stable decomposition results. User interactions are needed and a predefined image database must be provided. To overcome above shortages, a novel algorithm is proposed in the dissertation to decompose the skin pigment based on single image. The new method is entirely automatic and requires neither user interactions nor predefined image database. It's robust and can handle input images with poor resolution and noise very well. In addition, a new algorithm is presented to distinguish the melanin and hemoglobin distribution maps from the decomposed components. After a successful decomposition, several methods for analyzing the distribution of pigment concentration are introduced. These methods are proven to be very helpful in many applications.4) Novel algorithms to detect and remove highlightsSpecular reflection on object surface will induce local highlights in image, which will lead to incorrect results both on skin pigment separation and general image processing. A new concept called Best Fit Window (BFW) is introduced in Chapter 4 to obtain diffuse chromaticity and then to detect highlight regions. Furthermore, a new algorithm is developed to estimate illuminant chromaticity by linear fitting approach. Finally, STD mechanism is applied to successfully remove highlights from detected highlight regions based on estimated illuminant chromaticity.5) Basic researches on e-cosmeticE-cosmetic is a very new research direction. Based on the independent pigment distribution maps obtained, Chapter 5 introduces a new algorithm to remove freckle by applying texture synthesis technology. Chapter 6 then introduces methods to reproduce skin image by setting new pigment concentration and changing the color of skin image by applying color transfer technology. All the algorithms in these two Chapters can be considered as functions for e-cosmetic, which means do "make-up" digitally by using software tools.