classification - Feature Extraction for Face Dectection -
i'm required design (extract) 8 features collection of face images (see url below).
http://faculty.ucmerced.edu/mhyang/face-detection-survey_files/face-sample.gif
these images results of moving window (of fixed size) on number of original images. train naive bayes classifier using training set contains values of extracted features , class label.
so features should extract images? can give me examples?
a simple approach use pixel statistics, namely, mean , standard deviation (sd), of raw pixel values (assuming these greyscale values) of given region of face image (e.g. rectangular region containing top or bottom half of face).
if extract mean , sd of 4 different regions in each image, e.g., 2 rectangular regions , 2 circular regions, gives 8 numeric "high level" features.
the mean , sd represent pixel contrast in region, , "high level" features because cover group/area of pixels (whereas low level features use some/all of raw pixel values directly features).
see following research article details:
mengjie zhang, urvesh bhowan, “program size , pixel statistics in genetic programming object detection”. evolutionary computation in image analysis , signal processing. lecture notes in computer science. vol. 3005, 2004. pp. 377-386.
Comments
Post a Comment