Apparatus and method for detecting body parts

摘要

Provided is an apparatus and method for detecting body parts, the method including identifying a group of sub-images relevant to a body part in an image to be detected, assigning a reliability coefficient for the body part to the sub-images in the group of sub-images based on a basic vision feature of the sub-images and an extension feature of the sub-images to neighboring regions, and detecting a location of the body part by overlaying sub-images having reliability coefficients higher than a threshold value.

主权项

1. A body part detecting method, comprising:
generating a plurality of sub-images based on an input image comprising body parts; determining, by way of a processor, a reliability coefficient for the sub-images based on a basic feature of the sub-images and an extension feature of the sub-images to neighboring regions; detecting the body parts based on the determined reliability coefficient for the sub-images, wherein when it is determined that at least two of the sub-images have reliability coefficients higher than a threshold value among the sub-images, detecting the body parts by clustering is performed; and overlaying the sub-images by one of a direct overlay and a weighted overlay, wherein the threshold value is different based on a type of a multi-part context descriptor selected and combinations of multi-part context descriptors, wherein the determining comprises determining the reliability coefficient for the sub-images based on a similarity between the multi-part context descriptor for the sub-images and a preset multi-part context descriptor for the body parts, and wherein the multi-part context descriptor comprises a basic descriptor corresponding to the basic feature and an extension descriptor corresponding to the extension feature.