FACE RECOGNITION BY PIXEL INTENSITY

Transcription

1 FACE RECOGNITION BY PIXEL INTENSITY Preksha jain & Rishi gupta Computer Science & Engg. Semester-7 th All Saints College Of Technology, Gandhinagar Bhopal. Abstract Face Recognition is the inherent capability of human beings. Identifying a person by face is one of the most fundamental human functions since time immemorial. Face recognition by computer is to endow a machine with capability to approximate in some sense, a similar capability in human beings. To impart this basic human capability to a machine has been a subject of interest over the last few years. Such a machine would find considerable utility in many commercial transactions, personal management, security and the law enforcement applications, especially in criminal identification, authentication in secure system etc. Enough research has not been carried out on identification of human faces. However a number of automated or semiautomated recognition studies have been reported. Introduction:- Face recognition by pixel intensity, it is method in which there is image of any person taken by a camera. How our system will recognized that image. The first, and probably most significant challenge faced was transferring the images obtained from the camera to the computer. Two issues were involved with this challenge:- 1) Capturing and transferring in real time, 2) Having access to where the image was being stored. Human face localization and detection is often the first step in applications such as video surveillance, human computer Interface, face recognition and /or facial expressions analysis, and image database management. A lot of research has been done in the area of human face detection. In prior studies, different human skin colors from different races have been found to fall in a compact region in color spaces. Therefore skin can be detected by making use of this compactness. The face detection is performed in three steps. The first step is to classify each pixel in the given image as a skin pixel or a non-skin pixel. The second step is to identify different skin regions in the skin-detected image by using connectivity analysis. The last step is to decide whether each of the skin regions identified is a face or not. After the probable location of the face is found the left and the right edges of the face is determined. In the complete image how a camera will detect the face of the person. To do this we use pixels intensity. To use the method there are some algorithms that will apply on it. Face recognition system is used in various fields. The analysis of face images is a popular research area with applications such as face recognition, virtual tools, and human identification security systems. Design Issues The most important aspect of implementing a machine vision system is the image acquisition. Any deficiencies in the acquired images can cause problems with image

2 analysis and interpretation. Examples of such problems are a lack of detail due to insufficient contrast or poor positioning of the camera: this can cause the objects to be unrecognizable, so the purpose of vision cannot be fulfilled. Illumination A correct illumination scheme is a crucial part of insuring that the image has the correct amount of contrast to allow to correctly processing the image. In case of the face detection system, the light source is placed in such a way that the maximum light being reflected back is from the face. The face will be illuminated using a 60W light source. To prevent the light source from distracting the image, an 850nm filter is placed over the source. Since 850nm falls in the infrared region, the illumination cannot be detected by the human eye, and hence does not agitate the image. Since the algorithm behind the eye monitoring system is highly dependant on light, the following important illumination factors to consider are :- 1. Different parts of objects are lit differently, because of variations in the angle of incidence, and hence have different brightness as seen by the camera. 2. Brightness values vary due to the degree of reflectivness of the object. 3. Parts of the background and surrounding objects are in shadow, and can also affect the brightness values in different regions of the object. 4.Surrounding light sources (such as daylight) can diminish the effect of the light Source on the object. Camera Hardware The next item to be considered in image acquisition is the video camera. Review of several journal articles reveals that face monitoring systems use an infrared-sensitive camera to generate the eye images [3],[5],[6],[7]. This is due to the infrared light source used to illuminate the driver s face. CCD cameras have a spectral range of nm, and peak at approximately 800nm. The camera used in this system is a Sony CCD black and white camera. CCD camera digitize the image from the outset, although in one respect that signal amplitude represents light intensity the image is still analog. Frame Grabbers and Capture Hardware The next stage of any image acquisition system must convert the video signal into a format, which can be processed by a computer. The common solution is a frame grabber board that attaches to a computer and provides the complete video signal to the computer. The resulting data is an array of grayscale values, and may then be analyzed by a processor to extract the required features. Two options were investigated when choosing the system s capture hardware. The first option is designing a homemade frame grabber, and the second option is purchasing a commercial frame grabber. Light Source For conditions when ambient light is poor (night time), a light source must be present to compensate. Initially, the construction of an infrared light source using infrared LED was going to be implemented. It was later found that at least 50 LEDs would be needed so create a source that would be able to illuminate the entire face. To cut down cost, a simple desk light was used. Using the desk light alone could not work, since the bright light is blinding if looked at directly, and could not be used to illuminate the face. However, light from light bulbs and even daylight all contain infrared light; using this fact, it was decided that if an infrared filter was placed over the desk lamp, this would protect the eyes from a strong and distracting light and provide strong enough light to illuminate the face. A wideband infrared filter was placed over the desk lamp, and provides an excellent method of illuminating the face. The prototype of the Face Detection System is shown in Figure (a).

3 An explanation is given here of the face detection procedure. After inputting a image, pre-processing is first performed by binarizing the image. The top and sides of the face are detected. Using the sides of the face, the centre of the face is found, moving down from the top of the face, horizontal averages (average intensity value for each y-coordinate) of the face area are calculated. All images were generatingin Mat lab using the image processing toolbox. Binarization Binarization is converting the image to a binary image. Examples of binarized images are shown in Figure (c) Figure (a): Photograph of face Detection System prototype. System Flowchart A flowchart of the major functions of the Face Detection System is shown in Figure (b). 1. Threshold value Threshold value Threshold value 200 Figure (b) Figure (c): Examples of binarization using different thresholds.

4 A binary image is an image in which each pixel assumes the value of only two discrete values. In this case the values are 0 and 1, 0 representing black and 1 representing white. With the binary image it is easy to distinguish objects from the background. The grayscale image is converting to a binary image via thresholding. The output binary image has values of 0 (black) for all pixels in the original image with luminance less than level and 1 (white) for all other pixels. Thresholds are often determined based on surrounding lighting conditions, and the complexion of the face. After observing many images of different faces under various lighting conditions a threshold value of 150 was found to be effective. The criteria used in choosing the correct threshold was based on the idea that the binary image of the face should be majority white, allowing a few black blobs from the eyes, nose and/or lips. Figure (c) demonstrates the effectiveness of varying threshold values. Figure (c) 1, (c) 2, and (c) 3 use the threshold values 100, 150 and 200, respectively. Figure (c) 2 is an example of an optimum binary image for the eye detection algorithm in that the background is uniformly black, and the face is primary white. This will allow finding the edges of the face, as described in the next section. Face Top and Width Detection The next step in the eye detection function is determining the top and side of the face. This is important since finding the outline of the face narrows down the region in which the eyes are, which makes it easier (computationally) to localize the position of the eyes. The first step is to find the top of the face. The first step is to find a starting point on the face, followed by decrementing the y-coordinates until the top of the face is detected. Assuming that the person s face is approximately in the centre of the image, the initial starting point used is (100,240). The starting x-coordinate of 100 was chosen, to insure that the starting point is a black pixel (no on the face). The following algorithm describes how to find the actual starting point on the face, which will be used to find the top of the face. 1. Starting at (100,240), increment the x- coordinate until a white pixel is found. This is considered the left side of the face. 2. If the initial white pixel is followed by 25 more white pixels, keep incrementing x until a black pixel is found. 3.Count the number of black pixels followed by the pixel found in step2, if a series of 25 black pixels are found, this is the right side. 4. The new starting x-coordinate value (x1) is the middle point of the left side and right Side. Figure (d) demonstrates the above algorithm. Starting point (100,240) A Count 25 white pixels from this point B Count 25 black pixels from this point X1 Middle point of A and B Figure (d): Face top and width detection. Using the new starting point (x1, 240), the top of the head can be found. The following is the algorithm to find the top of the head: 1. Beginning at the starting point, decrement the y-coordinate (i.e.; moving up the face). 2. Continue to decrement y until a black pixel is found. If y becomes 0 (reached the top of the image), set this to the top of the head. 3. Check to see if any white pixels follow the black pixel. i. If a significant number of white pixels are found, continue to decrement y. ii. If no white pixels are found, the top of the head is found at the point of the initial black pixel.

5 Once the top of the driver s head is found, the sides of the face can also be found. Below are the steps used to find the left and right sides of the face. 1. Increment the y-coordinate of the top (found above) by 10. Label this y1 = y + top. 2. Find the centre of the face using the following steps: i. At point (x1, y1), move left until 25 consecutive black pixels are found, this is the left side (lx). ii. At point (x1, y1), move right until 25 consecutive white pixels are found, this is the right side (rx). iii. The centre of the face (in x-direction) is: (rx lx)/2. Label this x2. 3. Starting at the point (x2, y1), find the top of the face again. This will result in a new Y-coordinate, y2. 4. Finally, the edges of the face can be found using the point (x2, y2). i. Increment y-coordinate. ii.move left by decrementing the x- coordinate, when 5 black consecutive pixels are found, this is the left side, add the x- coordinate to an array labeled left_x. iii.move right by incrementing the x- coordinate, when 5 black consecutive pixels are found, this is the right side, add the x- coordinate to an array labeled right_x. iv. Repeat the above steps 200 times (200 different y-coordinates). The result of the face top and width detection is shown in Figure (e), these were marked on the picture as part of the computer simulation. As seen in Figure(e), the edges of the face are not accurate. Using the edges found in this initial step would never localize the eyes, since the eyes fall outside the determined boundary of the face. This is due to the blobs of black pixels on the face, primarily in the eye area, as seen in Figure(c) 2. To fix this problem, an algorithm to remove the black blobs was developed. Removal of Noise The removal of noise in the binary image is very straightforward. Starting at the top, (x2,y2), move left on pixel by decrementing x2, and set each y value to white (for 200 y values). Repeat the same for the right side of the face. The key to this is to stop at left and right edge of the face; otherwise the information of where the edges of the face are will be lost. Figure (f) shows the binary image after this process. Figure (f): Binary picture after noise removal. Figure(e): Face edges found after first trial. After removing the black blobs on the face, the edges of the face are found again. As seen below, the second time of doing this results in accurately finding the edges of the face.

6 The valleys are found by finding the change in slope from negative to positive. And peaks are found by a change in slope from positive to negative. The size of the valley is determined by finding the distance between the peak and the valley. Once all the valleys are found, they are sorted by their size Figure (g): Face edges found after second trial. Finding Intensity Changes on the Face The next step in locating the eyes is finding the intensity changes on the face. This is done using the original image, not the binary image. The first step is to calculate the average intensity for each y coordinate. This is called the horizontal average, since the averages are taken among the horizontal values. The valleys (dips) in the plot of the horizontal values indicate intensity changes. When the horizontal values were initially plotted, it was found that there were many small valleys, which do not represent intensity changes, but result from small differences in the averages. To correct this, a smoothing algorithm was implemented. The smoothing algorithm eliminated and small changes, resulting in a more smooth, clean graph. After obtaining the horizontal average data, the next step is to find the most significant valleys, which will indicate the eye area. Assuming that the person has a uniform forehead (i.e.; little hair covering the forehead), this is based on the notion that from the top of the face, moving down, the first intensity change is the eyebrow, and the next change is the upper edge of the eye, as shown below. Results and Future Work Simulation Results The system was tested on 15 people, and was successful with 12 people, resulting in 80% accuracy. Figure (h) (i) below shows an example of the step-by-step result of finding the eyes. Figure(h): Demonstration of first step in the process for finding the eyes original image. Figure (i): Demonstration of second step binary image.

7 Figure (L) Demonstration of step 5 - second edge detection. Notice that the second time the edges is found accurately. Figure (j): Demonstration of third step initial edge detection image. Notice that initially the edges are not found correctly. Figure (k): Demonstration of the fourth step - removing blobs in binary image. Figure (m): Result of final step - finding the left eye using intensity information. Limitations With 80% accuracy, it is obvious that there are limitations to the system. The most significant limitation is that it will not work with people who have very dark skin. This is apparent, since the core of the algorithm behind the system is based on binarization. For dark skinned people, binarization doesn t work. The most important aspect of implementing a machine vision system is the image acquisition. Any deficiencies in the acquired images can cause problems with image analysis and interpretation. Examples of such problems are a lack of detail due to insufficient contrast or poor positioning of the camera: this can cause the objects to be unrecognizable, so the purpose of vision cannot be fulfilled. Future Work Using 3D images is another possibility of finding the face. Adaptive binarization is an addition that can help make the system more robust. This may also eliminate the need for the noise removal function, cutting down the computations needed to find the face. This will also allow adaptability to changes in ambient light. The system does not work for dark skinned individuals. This can be corrected by having an adaptive light source. The adaptive light source would measure the

8 amount of light being reflected back. If little light is being reflected, the intensity of the light is increased. Darker skinned individual need much more light, so that when the binary image is constructed, the face is white, and the background is black. Application Areas Toy - Intelligent robotic. Casino-Filtering suspicious gamblers / VIPs. Vehicle-Safety alert system based no eyelid movement. Immigration/Customs-illegal immigrant detection. Passport / ID card authentication Government Events -Criminal/Terrorists screening; Surveillance Enterprise Security- Computer and physical access control. Bibliography [1] Davies, E.R. Machine Vision: theory, algorithms, and practicalities, Academic Press: San Diego, [2] Dirt Cheap Frame Grabber (DCFG) documentation, file dcfg.tar.z available from ook/video/ [3] Gonzalez, Rafel C. and Woods, Richard E. Digital Image Processing, Prentice Hall: Upper Saddle River, N.J., [4] Perez, Claudio A. et al. Face and Eye Tracking Algorithm Based on Digital Image Processing, IEEE System, Man and Cybernetics 2001 Conference, vol. 2 (2001), pp [5] A Robust Skin Color Based Face Detection Algorithm Tamkang Journal of Science and Engineering, Vol. 6, No.4, pp (2003) 227,Sanjay Kr. Singh1, D. S.Chauhan2,MayankVatsa3,RichaSingh3*, Department of IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.3, March [6] Detecting faces in images: A survey, M.Yang, D.J.Kriegman and N.Ahuja, IEEE Trans. Pattern Analysis and achine Intell.24, 2002, pp [7] Face detection in color images.r.l.hus, M.A.Mottaleb and A.K.Jain,IEEE Trans. Pattern Analysis and Machine Intell. 24, 2003, pp [8] A New Color Transformation For Lips Segmentation.N. Eveno, A. Caplier and P.Y. Coulon,. InProc. MMSP 01, pp. 3-8, 2001.