Docstoc

Face Detection based on the feature Analysis against skin color like backgrounds

Document Sample
Face Detection based on the feature Analysis against skin color like backgrounds Powered By Docstoc
					National Conference on Role of Cloud Computing Environment in Green Communication 2012                                         559


                    Face Detection based on the feature Analysis against skin color like backgrounds
        1                                                      2
            Dr. C. Seldev Christopher                              R.Dhivya
        1                                                      2
            Professor                                              PG Student
        St.Xavier catholic college of Engineering              St.Xavier catholic college of Engineering
        cseldev@gmail.com                                      dhivya.17a@gmail.com

        Abstract: In this paper, we detect the face based on the facial features against the skin color-like
        backgrounds. This method consists of three stages. It includes Image processing, skin color segmentation,
        Determination of face candidate .In first step, the object in the foreground is separated from background.
        Second step, to reject the non skin color clusters. Third step, both the elimination of skin color like blobs in
        backgrounds and face detection is performed. In addition, it detects the face images in different pose and
        various expressions.
        Keywords-Face detection, Edge detection, Skin color segmentation.
               1.    INTRODUCTION
        In Face detection, they are numerous researchers due to its wide variety of application such as human computer
        interface (HCI), video surveillance, content based image retrieval. During the past decade, there are some effective
        methodologies to improve the performance of detection rate. However, the speed of face detection is still in problem
        it from being widely used in real world application. This problem is not completely solved until the algorithm of
        face detection based on AdaBoost and Haar like feature were proposed by viola and jones [1]. The color-based
        approaches are difficult to detect the skin-tone color in the presence of skin color-like backgrounds.
                     Normally, face detection based on skin-tone colors is affected by illumination variations and complex
        backgrounds. To overcome these drawbacks, the adaptive method is introduced to choose the optimal one from the
        nine combinations of skin color models by a well defined quality measures. The AdaBoost method with Haar like
        features to determine whether the human face exist in the cluster or not. This method is used to trace the images
        from dim light, dark light and bright light settings. To overcome the drawback of above, proposed the three stages
        mainly based on skin color and facial features. Although there is one advantage, the detection time is only 15.9ms
        for each frame. But the drawback is, in this model need more parameters to be tuned for robustly detecting possible
        face candidates. This method cannot well operate in cluster of human skin color overlapped in the skin color like
        backgrounds.
                     To overcome these drawbacks, three stages is proposed it include image processing, skin color
        segmentation, Determination of face candidate. The main advantage is, it remove the human skin color like blobs in
        the backgrounds. But the drawback is, it have high computational cost.
   2.   EXISTING SYSTEM
        The face detection including three stages of image processing, skin color segmentation, Determination of face
        candidates. The description of three stages are given below
   a.   Image processing




 Department of CSE, Sun College of Engineering and Technology
National Conference on Role of Cloud Computing Environment in Green Communication 2012                                          560


                 To speed up the image processing the input image is resized as 300X225 pixels .The input image is fixed as
        300 pixels .Based on the aspect ratio of input images the length of other side is obtained. Reference white is
        obtained by using the method of lighting compensation. The reference white uses 5% of luma values in the image .
                 Convert the RGB color space into YCbCr color space. The YCbCr color space independent on the luma it
        give better performance in various lighting conditions. The Y-color space obtained the gray image. The formula of
        YCbCr color space is obtained below
                 Y=0.299R+0.586G+0.114B
                 Cb=-0.169R-0.331G+0.499B
                 Cr=0.499R-0.418G-0.0813B
        To separate the foreground and background from the images by using edge detection and fill operation. Sobel edge
        detection method is used to detect the edges from the image. Those object points are enclosed to the edges are filled
        by white pixels by using the fill operation. Using 3x3 mask to perform the fill operation. The AND operation
        segment the object cluster from the backgrounds.
                 The 3x3 mask perform the fill operation, if the center point in the mask as black pixel and more than 4
        white points included except the center point then the gray level of center point is set as white pixel. This method
        scan the image from top to bottom and left to right to obtain the resulting image X. At the same time it scan the
        image from top to bottom and right to left to obtain the resulting image Y. Then perform AND logical operation for
        both the images, those points are enclosed to the edge points are filled by white pixels by using fill operation
   b.   Skin color segmentation
        The object clusters are separated from the backgrounds by using the image processing method. To eliminate the non
        skin object are proposed by Garcia and Tziritas [5]. The following equations are used to obtained the Y-space
                 If(Y 128)
                            2
                        20
                        6
                        8
                 If(Y   128)
                        6
                        12

                        2

                                16

        And
                 Cr≥-2(Cb+24), Cr≥-(Cb+17)
                 Cr≥-4(Cb+32), Cr≥2.5 (Cb+               ,
                 Cr         ,            0.5   ! ,
                            "#$%
                                     ,               !



 Department of CSE, Sun College of Engineering and Technology
National Conference on Role of Cloud Computing Environment in Green Communication 2012                                               561


        By using the equation (1) to determine the parameter from       '(     the values from Cb space restrict the Cr space of
        skin –tone color pixels. These methods eliminate the non skin color object but the skin color like blobs in
        background are still in the images.
   c.   Determination of face candidates
        This method eliminate both the skin color like blobs in background and skin color clusters without faces such as
        arms, hand, leg and so on. After the skin color segmentation, the morphological opening and erosion operations is
        performed. It separate the skin color like blobs in backgrounds and skin color object cluster. It completely split the
        two connected regions. 8 connected component labeling method is used to detect the connected regions in the image.
                 The skin color cluster includes the face part with neck and the non skin region without faces such as leg,
        arm, hand and so on. After performing the morphological opening and erosion operation the candidate face region is
        determined. By using the 8 connected component labeling method to draw the red rectangle box in the skin regions.
        It includes the face region and hand. The human faces are approximately assumed as elliptical, the histogram appear
        a single mode of Gaussian distribution.
                 The skin color cluster may not be contain any faces should be eliminated. Since the aspect ratio faces of 4/3
        is assumed, it indicate that if the width of skin color clusters is larger than its height, the skin color may be an arm ,
        or hand should be rejected. The clusters containing the face candidate say Wr , can be fitted to an ellipse say We
        shown in the fig 1




                                                -Wr


                                              -We




        The face candidate accepted by elliptical fitting may contain no faces, and these face candidates should be further
        removed [1].




            1.   PROPOSED WORK
        The main drawback of the existing system is it does not support if the illumination is low. So a new lightening
        compensation technique is proposed to support the low illumination.
                 4. CONCLUSION



 Department of CSE, Sun College of Engineering and Technology
National Conference on Role of Cloud Computing Environment in Green Communication 2012                                  562


      In this paper, the face detection method against skin color like backgrounds has been successfully presented.
      Moreover, higher performance of detection rate can be achieved even though faces are different poses or
      expressions.
      REFERENCES
      [1] P. viola, and Mr.Jones, “Rapid object detection using a Boosted cascade of simple features” In proc IEEE
      conference of computer vision and pattern recognition, Hawaii , USA 2001.
      [2] R.L. Hsu, M. Abdel-Mottaleb, and A.K. Jain, “Face detection in color image,” IEEE Transactions on Pattern
      Analysis and Machine Intelligence, Vol. 24, No. 5, 2002, pp. 696-706.
      [3] D.Y. Huang, W.C. Hu, and M.H. Hsu, “Adaptive skin color model switching for face tracking under varying
      illumination,” In: Proc. IEEE 4th International Conference on Innovative Computing, Information and Control
      (ICICIC2009), Kaohsiung, Taiwan, 2009, pp. 326-329.
      [4] W.C. Hu, C.Y. Yang, D.Y. Huang, and C.H. Huang, “Real-time and reliable face detection in intersection
      monitoring by integration of skin color and facial features,” In: Proc. IEEE 4th International Conference on
      Innovative Computing, Information and Control (ICICIC2009), Kaohsiung, Taiwan,
      2009, pp. 1160-1163.
      [5] C. Garcia, and G. Tziritas, “Face detection using quantized skin color regions merging and wavelet packet
      analysis,” IEEE Transactions on Multimedia, Vol. 1, No. 3, 1999, pp. 264-277.
      [6] D.Y. Huang, and C.H. Wang, “Optimal multi-level thresholding using a two-stage Otsu optimization approach,”
      Pattern Recognition Letters, Vol. 30, No.3, 2009, pp. 275- 284.
      [7] P.J. Phillips, H. Moon, P.J. Rauss, and S. Rizvi, “The FERET evaluation methodology for face recognition
      algorithms,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 22, No. 10, 2000, pp. 1090-
      1104.
      [8] D.Wu-Chih Hu, “Face detection based on feature analysis and edge detection against skin color like
      backgrounds”, IEEE conference on Innovative computing ,2010.




 Department of CSE, Sun College of Engineering and Technology

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:24
posted:7/26/2012
language:English
pages:4