| Summary: | 碩士 === 國立中興大學 === 資訊科學系所 === 94 === Face detection and feature location are the pre-processing of many facial researches having the condition of limit face size and the environment to detect the faces.
Our proposed method is able to detect face in a complex environment and light conditions, even if the resolution of the image is low.
For the part of face detection, we use rectangle features and integral image to instead of using traditional pixel based methods.
In this method, we use Adaboost algorithm to obtain rectangle features for different face.
A different weight can be updated automatically to classify the feature of face by using this algorithm.
In order to keep the face-like region, we apply the method of cascade in which the background region can be discarded more quickly to separate with the face-like region.
In the part of feature detection, we identify the region of face by using previous steps and locate the organs of face via the relative position.
Then, we can get the exact point of feature based on the color and edge information.
To reduce the error rate of identification, the character of moving image can be used here as a reference in which the continuous frame having the feature of slight change within 5fps in general.
Thus, we can discard the diverse frame and replace it with the previous five frames.
|
|---|
