A low-level approach to semantic classification of mobile multimedia content
Anand Mariappan, Michael Igarta, Cuneyt M. Taskiran, B. Gandhi, Edward J. Delp · 2005
Pseudo-semantic labeling is a novel approach to organize mobile multimedia content such as images and videos. We have developed low-complexity algorithms to derive labels, such as "indoor/outdoor", "face/not face", that can be run on the mobile device. "Indoor/outdoor" classification is done based on the presence of sky in the images. Skin like pixels are detected based on the color information and if present they are matched to a pre-defined "facetemplate" to detect the presence of faces in images. We have developed an initial framework for finding the degree of blurriness in an image, which can be used for labels like "blurry/not blurry".