An Implementation Of Face Detection And Recognition System Based On Images Collected Naturally

As the development of Computer Science, Facial Recognition gets more attention. There have been a great number of advances in this field in the recent years; AdaBoost broached by Intel is a representative among them. In this treatise, I study and analysis this algorithm intensively and improve it to combine it with the traditional recognition algorithm. I use this technique in naturally collected pictures and get some positive effects while maintaining the origin algorithm's real time property. With an experiment on members of our lab, the accuracy and the effectiveness of our system got verified. This experiment contains the following aspects.(1) Extract a human face from a naturally collected picture. In the traditional facial recognition system, the gallery comprises of artificial pictures made exclusively for face recognition. In our system we can detect the human faces from the natural picture by using the AdaBoost technique, giving face recognition more opportunities for wider application.(2) Processing the prior detected faces. The faces got from prior step have background associated with it. If handled not well, this background may have an ill impact on the later face recognition step. So in our system, we try to remove this background to improve our system's accuracy.(3) Classify the detected faces by using the Eigenface method. The most difficult thing of using this technique is to choose the number of feature vector. In this experiment we have a number of trial to find the sub-optimal number to balance between the real time requirement and the accuracy.(4) Post-process the selected faces. When using the pattern classification technique, sometimes you won't the only result. You should have a number of post processing step in order to get the result that best meets our requirements. In our system, we use the KNN (K Nearest Neighbors) method to get the optimal result.(5) Implement the system. This system finally got implemented by using Java and C++. When the input is the standard surveillance video stream, this system can reach the real time requirement. So this system got some application value.