International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 14 - Mar 2014 Simple Face Detection Using Skin, Hair and Edge Characteristic Mukta K.Dutonde#1, Lokesh Bijole*2 # * Student, Computer Engineering, Padm. Dr. V.B. Kolte COE, Malkapur, India Assistant Professor, Computer Engineering, Padm. Dr. V.B. Kolte COE, Malkapur, India Abstract-This article presents two different algorithms for rapid and accurate face detection: The decision algorithm and edge tracking algorithm. The decision algorithm based on skin and hair color characteristics and decision structure which converts the obtained information from skin and hair regions to labels for identifying the object dependencies and rejecting many of the incorrect decisions. The edge tracking algorithm used to extract the sub windows from the image to extract the rectangle features while skipping the backgrounds. Keywords—Edge detection, label, object dependencies, threshold, skin region. I. INTRODUCTION The human Face is one of the most important objects of computer vision. It has a wide range of applications in biometrics recognition, computer vision and multimedia applications. Detection of face is quite a complex procedure and involves the feature extraction, based on which face can be detected. There is several approaches for human face detection, which are categorized into four types (1) Knowledge-based method: This method finds invariant features of a face and localizing the position of face. The features are used to check whether human face is appearing in an ISSN: 2231-5381 image or not [1]. (2)Feature invariant method: The method involves facial feature that are invariant to pose, brightness and viewpoints. Skin colors [2] fall into this category. (3)Template Matching: This method calculates Correlation between a test image and pre-selected facial templates [3].(4)Appearance based method: This method uses machine learning techniques to extract discriminative feature from a relabeled training set. The Eigen faces method [4] and AdaBoost-based face detection [5] is belonging to this class. In this paper the two different methods and the advantages of their algorithm is discussed. The algorithm based on knowledge based method adopts characteristics of skin and hair colors [6].while the algorithm based on appearance based method adopts characteristics of edges [7]. This paper is organized as section 2 presents Decision Algorithm based on skin and hair color, section 3 presents the edge tracking algorithm, section 4 presents the performance analysis, section 5 presents the conclusion. II. DECISION ALGORITHM The input image is divided into two parts which are determined by the conditions of http://www.ijettjournal.org Page 724 International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 14 - Mar 2014 the skin and hair color regions. Depending on the chosen color area, we can use Different methods for skin detection. The intensity element in RGB model allows us to find a suitable region for dark color hairs. The outcomes from here will be segmented by 5×3 or 5×5 blocks. Hereafter, we use edge detection with the “Canny” filter to recognize the external and marginal Edges. This filter plays an important role in the detection procedure because it preserves the edge continuity better in comparison with other filters such as “Sobel” and “Prewitt”. Then each result matrix will be labeled and converted to a matrix of objects. Then algorithm makes decision which of the labeled objects has correlation with others. The performance of the algorithm is explained in the next parts in details. Fig. 1 exhibits the total view used here for recognition. labels.3)[y] Hair+3:It is same as that certain hair labels matrix, when shift all of it’s ‘Y’ components by 3 pixels. The objects having scale less than THS or THH are considered as illegible for face locations and need to remove its information from the next consideration. After that it finds the intersection of skin and the matrix ‘[Y] Hair+3’.The condition to find out relationship between two labels are defined by two parameters that are Mysi and MYHj . The MYSi and MYHj statements are: (1) MYSi = (2) MYHj = Where [x] =max {nЄz/n≤x} or the floor of x. The condition MYSi-10> MYHj allows hair labels to proceed through the algorithm that are placed in a higher location. For each skin label, the eligibility of the whole Hair labels are examined and the other skin labels follow the same procedure. After examining all the skin and hair labels, it uses the Border and Center detection module to obtain the Border of any face in isolation. The formula is: Fig.1.Diagram of the system The decision algorithm [6] includes threshold for skin and hair labels as it extract the hair and skin characteristic for face detection.1) THS: The threshold for skin labels 2) THH: The threshold for hair ISSN: 2231-5381 Where Fi means the ith face detected and ‘~’ shows the assumed labels belong to Fi. Also, in the following equations, the mathematic symbol ‘|’ means “such that”. In continue, when we say for an example Fi X ∈ I or Fi Y ∈ I , we have assumed that Xor Y contains non-zero information. http://www.ijettjournal.org Page 725 International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 14 - Mar 2014 obtained mask to a new location without any change in its proportions, such that the previous center is placed in the location of the statistical center of the skin label matrix calculated as follows: XNew Center = YNew Center = Now the detection process is complete. III. EDGE TRACKING ALGORITHM We first concentrate on Preprocessing of images which include enhancement and feature extraction. A. Enhancement The input face images may have the control of lighting condition. So the histogram equalizations is used to compensate for lighting conditions and improve quality of image. Also the images contain the noise. The fine details of the image represent high frequencies which mix up with those of noise. The low pass filtering and medium filtering is used to remove the noise but in low pass filtering some details of image are also erased. So the median filtering is used to suppress the noise. In median filtering the value of a pixel is replaced by the median of the values in the neighborhood of the pixel. B. Feature extraction Fig.2.Decision Algorithm Flowchart To mask the detected faces center shifting is used. Then the effective features in determining the face location by this algorithm are the skin labels. With this assumption, we transfer the previous ISSN: 2231-5381 Here we are using the rectangle feature. There are tree three kinds of rectangle features the value of a two rectangle feature is the difference between the sums of the pixels within two rectangle regions. The regions have the same size and shape and are horizontally or vertically adjacent as shown in the figure. A three rectangle feature computes the sum within two outside rectangles subtracted from the sum in a http://www.ijettjournal.org Page 726 International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 14 - Mar 2014 center rectangle. Finally a four rectangle feature computes the difference between diagonal pairs of rectangles. These rectangle features are computed using an intermediate representation for the images which we call the integral image. Fig:3 Four types of rectangle features Edge tracking algorithm[7] eliminate the background noise and to reduce the number of sub-windows to be scanned while subsampling. window. In the same row, then it searches for the next edge that it is called as top-right coordinate. Then the algorithm searches in the column for the edge pixel. if it finds an edge pixel at any point in the column is considered as bottom-right & bottom-left coordinates using the above four coordinates, the sub-window is extracted from the edge image and their mean is calculated. The sub-window is considered as background if the mean value is zero and skipped that window to extract the next subwindow. If mean value is not equal to zero then the corresponding sub-window is extracted from the original image with the same coordinates and the rectangle features have been calculated. The following figure tracking algorithm: illustrates edge The algorithm is given as follows: Edge Tracking Algorithm 1. Load the edge image Eij 2. For each row 3. For each column i. Search for bottom-left, bottom-right top- right and top- left coordinates. ii. Extract the sub-window from the edge image. iii. Calculate its mean (m) iv. If m=0, then the sub-window is Considered as background. Goto Step 3 to extract the next subWindow. v. Else, extract the corresponding Sub window from the original image and Calculate the rectangle features. The edges are detected from the face image. In the edge image, the algorithm searches for first edge pixel in the each row. when it finds an edge pixel, the edge-pixel is considered as top-left coordinate of the sub ISSN: 2231-5381 Fig:4 Output from edge tracking Algorithm http://www.ijettjournal.org Page 727 International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 14 - Mar 2014 IV. PERFORMANCE ANALYSIS Face detection using skin and hair gives no guarantee that it detects only skin and hair pixels. Because the data from other parts of the image is also included with this color in it. So the decision algorithm is used to detect the face characteristics and reject other. In the decision algorithm, labeling technique is used to label the face elements i.e. hair and skin, to several labeled objects. Object scale comparability is the advantage of this algorithm. This property helps us to remove the small size objects by a fraction of the greatest object size with respect to a threshold determined arbitrarily. Another advantage is preserving the combinability with any other method of face recognition in any step to improve the reliability. Here we deplete the objects of the information matrices quickly with special filters such that the remained edge contains the same useful information as the primary matrix had before. That means the load of processing decreases and subsequently the speed of diction can increase. This algorithm has logic to determine the object dependencies. While in edge tracking algorithm almost all the systems are classifying the faces starting with some base size. Also the entire image is sub-sampled to identify the faces. To eliminate the background and to reduce the number of sub-windows to be scanned. While sub-sampling, thus it performance faster. From the above discussion we can say that the decision algorithm is better than edge tracking algorithm because the decision algorithm extract features of skin and hair from input image and then identify the edges to remove the unnecessary information from the image input. As seen before the edge tracking algorithm just try to find out the edges of image which is unable to remove the noise from the image and it also contain some ISSN: 2231-5381 unnecessary information that complexity during image detection. V. cause CONCLUSION In this paper, we determine the two different algorithms and their advantages. The decision algorithm is capable of detecting several human faces at the same time and sharing information with other methods to increase the overall system reliability as could be seen from the algorithm flowchart it uses simple logical operations on small amount of data which is the labeled objects to recognize the faces. The main advantage of the edge tracking algorithm is detecting the face with minimal number of scanning. REFERENCE [1] Jerome M. Shapiro, “Embedded Image Coding Using Zerotress of Wavelet Coefficients”, IEEE Transaction on Signal Processing Vol.41 No.12 December 1993. [2] J. Yang and A.Waibel, “A Real-Time Face Tracker” ,Proc. Third Workshop Applications of Computer Vision, 1996,pp. 142147. [3] Lanitis, A. Taylor, C.J. and Cootes, T.F.(1995). An automatic face identification System Using Flexible Appearance Models.Image and Vision Computing, vol.13,no.5, pp.393-401. [4] Turk, M.and Pentland, A. (2000). Face recognition using eigenfaces.Proc. IEEE Conference on Computer Vision and Pattern Recognition, pp. 586–591. [5] Viola, P. and Jones, M J. (2001).Rapid object detection using a boosted cascade of simple features. Proc. Conference On Computer Vision And Pattern Recognition 2001 (CVPR). [6] P. Ghofrani, Z. Neshat and H. Aghaeinia(2012).Labelling Algorithm For face detection using skin and hair Characteristics.Journal of Eletronic Science and Tech,Vol.10,NO.2 [7] S.Anila and N.Devarajan(2010).Simple And Fast face detection system based on Edges. International Journal of Universal Computer Sciences.Vol.1-2010/Iss.2. http://www.ijettjournal.org Page 728