Robust skin-roughness estimation based on co-occurrence matrix☆,☆☆
Introduction
Traditionally, research on human-skin analysis has been mainly conducted in the context of medical imaging [1], [2], [3], [4], [5]. Exclusive skin diagnostic equipment such as dermatoscopy has a magnifying glass to scrutinize the skin status [6], and it is designed to block outside light to acquire a skin image in a consistent environment [7]. Even though these special equipments accurately diagnose skin condition, it is troublesome and time-consuming. For example, people have to visit a dermatology or skin-care shop for skin analysis. In addition, they should put their faces into diagnostic devices to block external lights [8], [9]. It would be convenient and popular, even for people not interested in skin care, if a portable mobile device could provide skin-care service.
Mobile phones are widespread and have advanced significantly, particularly in terms of the imaging sensor. Moreover, they have the capability of acquiring a high-resolution image that can sufficiently capture skin textures [10]. In this paper, we propose a new technique to analyze the skin texture from an image taken by a mobile-phone camera, especially focusing on the skin roughness, which is one of major factors for skin-condition measurement.
The skin texture is characterized by very tiny variations in a signal intensity level; thus, its image is very vulnerable to noise and the surrounding illumination during sensing. The work in [6] tried to estimate the influence of the surrounding illumination on a skin image, and the work in [11] applied color constancy algorithms to dermatoscopy image classification for an illuminant environment.
To alleviate the illumination effect, we propose to use the texture domain, on which the skin texture is analyzed. A skin image can be decomposed into the sketchy and texture components. The former commonly corresponds to a blurred approximation of an input image, whereas the latter contains signal details [12], [13]. The texture component includes most of the background region of a skin image; thus, the skin roughness could be easily estimated.
Processing on the texture domain has two key benefits. First, the texture component decomposition can reduce the influence of the surrounding illumination because the slowly varying illumination effect in an image tends to be classified into the opposite sketchy component. The other benefit is that sketchy-texture decomposition can separate the major skin components such as the pores, hair, and moles as well as the illumination effect from the background skin texture. These skin components are not directly related to the skin roughness and may be an obstacle for the accurate skin-roughness estimation.
Skin-roughness estimation is closely related to texture analysis, which is one of the traditional research topics in image processing. There are many conventional texture-analysis methods, which are commonly classified into statistical, structural, and model-based categories [14]. Statistical methods extract the overall representative value from the image-signal distribution to classify the texture pattern. The structural methods include compactness, topological descriptors, and morphological approaches. The model-based approaches generally exploit Markov, pyramid, Gabor, wavelet, and Fourier transforms. Among these texture-analysis methods, we particularly choose to use the co-occurrence-matrix-based approach that belongs to the statistical category.
Co-occurrence matrices have been widely used for texture analysis for a long time and a number of co-occurrence features have been proposed in previous works. Since 14 features were proposed in [15], most previous texture-analysis works basically exploited these features to analyze a texture image [15], [16], [17], [18], [19]. Although these features exhibit satisfactory performance for classifying distinctive texture patterns, it is still challenging to analyze subtle textures such as the skin roughness that is considered in this study. Therefore, we identify the limits of conventional gray-level co-occurrence matrix (GLCM) feature-extraction methods for skin-roughness estimation and propose a new GLCM feature-extraction method that is especially useful for skin-texture analysis.
Fig. 1 shows the overall flowchart of the proposed skin-roughness estimation method. First, a skin image is acquired using a mobile-phone camera under various daily illumination environments. Since the influence of illumination is very critical in the analysis of tiny skin-texture signal, it should be alleviated for the accurate estimation of skin roughness. Moreover, some skin components such as pores, pigments, and moles that are less related to the skin roughness should be also excluded. Thus, we propose to estimate skin roughness on texture domain as shown in the 2nd image from the left in Fig. 1. As a result, both the illumination effect and skin components are avoided significantly in the analysis of texture domain. This decomposition is elaborated in detail in Section 3. Next, the uniformity of the skin-texture region is measured by the proposed co-occurrence matrix feature. We use the GLCM to measure the uniformity under the assumption that a skin image with large variations is perceived as coarse skin. The skin texture is finally graded by comparing the feature value with those of training images. Finally, the skin roughness is graded into the following three classes: coarse, moderate, and silky.
This paper is organized as follows. Related works on skin-condition estimation are presented in Section 2. In Section 3, we elaborate on the texture-domain decomposition of a skin image in detail, and then, Section 4 introduces the proposed feature extraction method based on co-occurrence matrix. Section 5 presents experimental results that show the superiority of the proposed method, and Section 6 concludes the paper.
Section snippets
Related works
There are several characteristics that represent skin conditions: wrinkles, pores, pigments, roughness, luster, and so on. In our previous works [20], [21], the number of pores is estimated by analyzing a skin image taken by a mobile-phone camera. The work in [20] presents methods that reduce the influence of the surrounding illumination and extract the pore region. In [21], the relationship between the shooting distance and the morphological parameters is found for denoising because the pore
Texture component extraction
In this work, the skin roughness is defined by the extent of the intensity variations in the skin-texture signals. When the skin roughness is graded, the skin texture itself should be evaluated, excluding other skin components such as pigmentation, hair, and pores in a skin image since these skin components actually has little relation with roughness. In addition, the skin image is acquired by a mobile-phone camera in a daily environment. This means that it is easily affected by variations in a
Gray-level co-occurrence matrix
The standard deviation and moments of the pixel intensities are common ways to measure the uniformity of an image. However, these measures do not reflect the certain properties of a signal pattern. They just represent a numerical quantity related to the overall distribution of image signals. Further, co-occurrence matrix indicates a 2D distribution that reflects the geometric relation between neighboring pixels. It is more appropriate for estimating the subtle changes in the texture component
Experimental results
To evaluate the proposed skin-roughness classification method, we acquire 100 close-up skin images, as shown in Fig. 5, using a Samsung Galaxy S4 zoom mobile phone in a daily environment. All skin images were taken without flash because it is difficult to capture skin roughness due to extreme illumination. The resolution of skin images taken for experiments is 16M and region is cropped in the center of the image. The distance between the mobile camera and the skin is fixed to 3 cm, and
Conclusions
In this paper, we propose a method that estimates the skin roughness using an image taken by mobile-phone camera to replace conventional skin diagnostic equipment. This professional equipment is not general since it is less accessible and inconvenient. Skin images taken by a mobile-phone camera in everyday environments are highly affected by various types of illumination. In addition, skin-roughness estimation is often hindered by skin components such as pores and moles, which are not basically
References (29)
- et al.
Exemplar-based segmentation of pigmented skin lesions from dermoscopy images
- et al.
A contactless biometric system using multiple hand features
J. Vis. Commun. Image Representation
(2012) - et al.
Discriminative common vector based finger knuckle recognition
J. Vis. Commun. Image Representation
(2014) - et al.
Brief review of invariant texture analysis methods
Pattern Recogn.
(2002) - et al.
Prediction of skin quality properties by different multivariate image analysis methodologies
Chemometr. Intell. Lab. Syst.
(2009) - et al.
Comparison of segmentation methods for automatic diagnosis of dermoscopy images
Eng. Med. Biol. Soc. (EMBS)
(2007) - et al.
A novel fusion approach for segmenting dermoscopy image based on region consistency
- P.G. Cavalcanti, Y. Yari, J. Scharcanski, Pigmented skin lesion segmentation on macroscopic images, in: Image and...
- et al.
Skin Roughness Assessment
(January 2010) - et al.
Characteristics of subjective recognition and computer-aided image analysis of facial erythematous skin diseases: a cornerstone of automated diagnosis
Br. J. Dermatol.
(2014)
Development and clinical validation of a novel photography-based skin pigmentation evaluation system: a comparison with the calculated consensus of dermatologists
Int. J. Cosmet. Sci.
Smartphones get even smarter cameras [future visions]
IEEE Consum. Electron. Mag.
Improving dermoscopy image classification using color constancy
J. Biomed. Health Infor.
Modeling textures with total variation minimization and oscillating patterns in image processing
J. Sci. Comput.
Cited by (8)
Gray level size zone matrix for rice grain classification using back propagation neural network: a comparative study
2022, International Journal of System Assurance Engineering and ManagementClassification of Basmati Rice Grains using Image Processing Techniques
2022, 2022 IEEE 6th Conference on Information and Communication Technology, CICT 2022Multi-modal Sensor and Data Processing for Comprehensive Skin Evaluation
2021, Proceedings of IEEE SensorsComparative analysis of texture feature extraction techniques for rice grain classification
2020, IET Image ProcessingPhotometric computer vision-aided system for psoriasis severity scoring: A preclinical study based on a mouse model of psoriasis
2020, Journal of Electronic Imaging
- ☆
This paper has been recommended for acceptance by M.T. Sun.
- ☆☆
This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2016-H8501-16-1017) supervised by the IITP (Institute for Information & communications Technology Promotion).