Detecting Noise Data in Face Recognition Using Geometric Moment Analysis Algorithm

Face detection and recognition is an effective and popular technique for authorizing the users and securing the data at the time of transactions. However, there exists some other traditional face recognition approaches with some disadvantages when opted to various types of input conditions. They are Geometric, Photometric and Principal Component Analysis etc.

The geometric approach deals with the geometric features that are generated by a sector, edges and regions of some shapes formed by many points. The recognition result is compared by analyzing the featured set. A calculation is done between the features in the template image and   every image in the database. Hence this method is tough. But the main complication is locating a point automatically. Also the other problem arises if the image is of poor quality.

On the other hand the photometric approach refines an image into values and that value is compared with templates to remove variances. It depends on the input image and the geometric location of different angles. While the photometric transformation is employed on the source image, it does not consider the photometric change which is nothing but the changes in the pixel. This approach requires multiple registered images of the same person. If any images which are not present in the dataset are subjected to processing, this approach considers that image as a new image which is specified as unauthenticated.

Principal Component Analysis is an orthogonal linear transformation which maps the data to another coordinates. It uses Eigen faces. This approach processes the images of equal sizes. Also this approach reduces the sizes of data using data compression technique. The images get disintegrated to form unrelated blocks that are stored in one dimension array know as Eigen faces. The face images can be signified as a sum of Eigen faces.

In order to overcome this incompetence, a new face recognition scheme based on invariant moment features which assure a secure transaction is proposed. Also the proposed scheme deals with an effective preprocessing using Short Time Fourier Transform (STFT), image enhancement techniques, extraction of local and global information using Region of Interest (ROI) calculation by the method of subdividing the determined ROI region into multiple sub ROIs. The ROI mainly focuses on the local features in face such as eyes, nose and lips. Face values in different angles were observed by calculating the area and centroid of the face using the above parameters which results in higher matching accuracy in the experimental results.