Hybrid Human Skin Detection Using Neural Network and K-Means Clustering Technique

Highlights

• The proposed method is a hybrid method that combines the neural networks and k-means.

• Modelling skin with different color-texture descriptors was investigated. High accuracy of F1 = 87.82% can be achieved based on images from the ECU database.

• Study showed that YIQ color space gives the highest separability in skin detection.

Abstract

Human skin detection is an essential step in most human detection applications, such as face detection. The performance of any skin detection system depends on assessment of two components: feature extraction and detection method. Skin color is a robust cue used for human skin detection. However, the performance of color-based detection methods is constrained by the overlapping color spaces of skin and non-skin pixels. To increase the accuracy of skin detection, texture features can be exploited as additional cues. In this paper, we propose a hybrid skin detection method based on YIQ color space and the statistical features of skin. A Multilayer Perceptron artificial neural network, which is a universal classifier, is combined with the k-means clustering method to accurately detect skin. The experimental results show that the proposed method can achieve high accuracy with an F₁-measure of 87.82% based on images from the ECU database.

Introduction

Separating an image into regions that consist of groups of identical linked pixels is an image processing stage called image segmentation. The homogeneity of a region can be defined by the color, gray levels, and texture, among other factors [1]. Skin detection is a good example of image segmentation, which can be accomplished by classifying image pixels as skin and non-skin pixels [2].

The importance of skin color detection comes from its use as a primary operation in many applications, such as face detection [3], surveillance systems [4], Internet pornographic image filtering [5] and gesture analysis [6]. For example, face detection is accomplished by taking out the joint facial characteristics and by employing skin color detection as a primary step to specify the face area. As a result, accurate and fast face detection can be accomplished.

Many researchers have investigated skin color features. Their results have shown that skin color has a limited range of hues and is not deeply saturated [7]. Thus, human skin color is clustered within a small area in the color space. In the past few years, several algorithms have been proposed for skin detection. However, some factors, such as illumination, may make skin color detection a very difficult task. [2]. The existing algorithms can be categorized into four classes: explicit skin classifiers, parametric classifiers, nonparametric classifiers, and dynamic classifiers [8]. The explicit classifiers, which are the easiest and are frequently employed, use a threshold strategy to distinguish between skin and non-skin pixels [9]. Basically, they characterize the limits of the skin region by utilizing a set of fixed thresholds. Although such classifiers are direct and might be used without any prior training steps, they may need adaptability when utilized under distinct imaging conditions. This may result in incorrect pixel detection [8].

Parametric classifiers can be based on a single Gaussian model [10], multiple Gaussian clusters [11], a mixture of Gaussian (MoG) models [12], or an elliptic boundary model [13]. Generally, the characterization speed of these classifiers is slow. In fact, they need to process each pixel individually. Additionally, these methods have low detection accuracy, as they rely on approximated parameters rather than authentic appropriate skin colors [8]. Furthermore, their performance varies depending on the utilized color space [14].

In the nonparametric classifiers, a set of training data is essential for estimating the statistical model of skin color distribution [15]. The advantages of these classifiers are quick training and skin distribution shape independence [14]. Nevertheless, such classifiers are not precise enough because of the requirement for an unbounded amount of training information, which makes them appropriate in a constrained scope of imaging conditions [8].

To overcome the generality of the previous static skin model classifiers, dynamic classifiers, which are based on artificial neural networks (ANN) and/or genetic algorithms, have been proposed [16]. The flexibility and ability of ANNs to adapt to various image conditions make them a good choice for enhancing classification tasks for human skin pixels [7].

Most of the existing skin detection methods segment images using only skin color information. Based on the fact that the skin region of an image is a group(s) of homogenous connected pixels, texture information can be used to describe skin regions. Different texture descriptors, such as homogeneity, uniformity, and standard deviation, may be exploited for detection purposes [17].

Although color information plays the main role in modeling of skin, selecting the prober color space to present skin is crucial. Several comparisons between different color spaces used for skin detection can be found in the literature [18], [19], [20], [21], [22], but one important question still remains unanswered is, “what is the best color space for skin detection?” Many authors do not provide strict justification of their color space choice. Some of them cannot explain the contradicting results between their experiments or experiments reported by others [22]. Moreover, some authors think that selecting a specific color space is more related to personal taste rather than experimental evidences [18].

In this paper, a hybrid classifier that combines ANN with k-means clustering is proposed. The proposed classifier exploits the color and texture information to detect skin regions. This paper is organized as follows. In Section 2, an overview of related skin detection methods is presented. Section 3 describes the existing algorithms that form the background for the proposed method. The proposed skin detection method is described in Section 4. The experimental results are reported in Section 5; finally, a research discussion is presented in Section 6, and the conclusion is summarized in Section 7.