| Human face recognition is a complex problem in computer vision and accuracy of an algorithm for face recognition depends on a lot of factors such as quality of input image, the accuracy of the algorithm, time complexity, space complexity, quality of training data, the accuracy of training algorithm etc. In our proposed process of face recognition, we use a much simpler approach that not only produces equally efficient output in low-quality and high-quality input images but also follows a simple image matching approach using facial symmetry features. First, we used Viola and Jones algorithm described in Viola & Jones (2004), to detect the face boundary which gives us some horizontal facial features such as ear line in both left and right side of the image and vertical facial features such as the top of the forehead and bottom of the chin. The rest of the features are extracted from the position of eye, nose and mouth. To detect the position of mouth, nose, and eye we devised an efficient algorithm that is capable of producing output with RGB, grayscale and B/W input image which makes this approach robust and portable. After detecting these facial land points, we calculate horizontal and vertical symmetry features. These symmetry features are extracted from both the reference and sample image and forwarded to our matching algorithm which gives us the percentage of similarity between two images. |
