![]() ![]() ![]() Thus, the 2-D image is transformed to K 2-D feature channels, where K is the number of feature channels or equivalently the dimension of the vector descriptor at each pixel (or a block of pixels). When HOG features are extracted from an image, blocks of pixels within the image are transformed to vectors. Of late, Histogram of Oriented Gradients (HOG) have been shown to perform well on a variety of detection tasks. The performance of the classifier is critically dependent on the choice of the feature representation. Discriminative feature representations in conjunction with features that generalize better than pixel values, can provide robustness against these challenges. However, pixel values (and to an extent edge maps) do not generalize well for object detection in unconstrained environments (e.g., street scenes, indoor scenes, etc.) due to background clutter and substantial variations in color, pose, etc. CFs, which have been extensively used for automatic target recognition (ATR) and biometric recognition, have traditionally been used with scalar features (usually raw pixel values or edge maps).
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |