For recognising the number and faces in a frame, should we with image based recognition or video based recognition approach. Please mention the cons and pros of both the techniques.
As for image based recognition, you may have this (possible) advantage: a high quality image available to be processed. You can use preprocessing steps which are averagely complex (from a computational time viewpoint). The problem is that a given face may be not recognized in the scanned image.
As for the video frames, the good thing is that, even if the algorithm cannot recognize the given face in a frame, it is likely that the algorithm can find other video frames in which the face can be recognized. So basically, you have many "available attempts". On the other hand, if you have to process many video frames (in real time), you can only use fast methods and the resolution of each frame cannot be too high.
I had a similar issue during the design/development of a barcode reader for mobile devices (decoding video frames from a video sequence vs asking the user to make a picture of the barcode to be decoded).
For face recognition, I applied neural networks, genetic algorithm, PCA and Gabor based approaches. Among them PCA (Principle Component Analysis) and Gabor based approaches produced better results. In order to make all faces with uniform lighting environment, used rms contrast scaling.