Image Reconstruction and Per-pixel Classification

Abstract

We describe face classification algorithm which can be used for object recognition, pose estimation, tracking and gesture recognition which are useful for human-computer interaction. We make use of depth camera (Creative Interactive Gesture Camera – Kinect®) to acquire the images which gives several advantages when compared over a normal RGB optical camera. In this paper we demonstrate a intermediate parsing scheme, so that an accurate per-pixel classification is used to localize the joints. We make use of an efficient random decision forest to classify the image which in turn helps to estimate the pose. As we employ depth camera to acquire depth image it may contain holes on or around depth map, so we first fill those holes and the classify the image. Simulation results was observed by varying several training parameters of the decision forest. We generally learned an efficient method which stems the basics in the development of pose estimation and tracking. Also we gained an intensive knowledge on Decision forests.