More than 3000 engineers find our updates useful. You can get them at your mail box!
Search your paper presentation and project titles:

Department/Area of interest: ( To list the projects / paper presentations)

Mechanical               Scada technology              Communication             Computer science           Alternative energy
Electrical                  Robotics                        Biometrics                     Artificial intelligence             Electronics

Automatic face detection using colour based segmentation and template/Energy threshold:

The project which is anticipated to give a huge advancement in technology deals with face detection based on color segmentation followed by template/energy matching. The drive of the project is to detect the presence and absence of face using computers. In EE368 (a graduate level introductory course to the fundamentals of digital image processing) they have given the task of training a system using collection of seven digital image for this face detection purpose to a set of students.

Here as the students face many complexities from deriving the algorithm for the successful execution of the project and to the simple steps to be followed, the first limitation was its time consumption of seven minutes for a single file. The students got catch hold of this concept not in their day one. They primarily started by reviewing all the articles on the topic and came up by trying out the methods that already existed such as neural networks, statistical methods, machine learning algorithms such as SVM,FLD,PLC(such as eigen faces and concept of “face space” as well as recent methodology such as “Maximum Rejection Classification”(MRC).

The students first tried out their hands with mainly two concepts “Eingen faces” and “neural networks”. They planned to have the Eigen faces based front end and neural network based back end. Eigen faces is a facial recognition technology that has set of Eigenvectors used in computer vision problem of human face recognition. But the neural networks were a great hindrance to clearly study the inner working of their system and were even unable to reveal the précised shortcomings in their system which may be rectified if found out. Thus they decided to leave this and thought of taking a different approach based on colour segmentation followed by template/Energy matching.

This approach has shown a better performance taking 80 to 120 seconds to run which relies on its internal down sampling rate applied to the input image and other parameters which can be altered. Its Performance accuracy was approximated to 85% to 100%. With their final parameters, in Dell 1.8 MHz Pentium IV laptop it takes about 100 seconds.

Download this complete project details here:

Intense Debate Comments