System and method for pet face detection

摘要

Pet faces can be detected within an image using a cascaded adaboost classifier trained to identify pet faces with a high detection rate and a high false positive rate. The identified potential windows are further processed by a number of category aware cascade classifiers, which are each trained to identify pet faces of a particular category with a high detection rate and low false positive rate. The potential pet faces identified by the category aware classifiers may be processed to verify whether a pet's face is present or not.

主权项

1. A method of detecting a pet's face in an image comprising:
receiving an image with one or more pet faces; processing the received image using a coarse cascade classifier trained to detect pet faces with a high detection rate and high false positive rate, the coarse cascade classifier identifying a plurality of potential pet face location windows within the received image; processing the plurality of potential pet face location windows to determine if a pet face is present, the processing of the plurality of potential pet face location windows comprising:
processing each of the plurality of potential pet face location windows with a first of a plurality of category-aware cascade classifiers trained on a set of pet images each associated with a first category, the processing by the first category-aware cascade classifier identifying zero or more of the plurality of potential face location windows as having a pet face present; andprocessing each of the potential pet face location windows not identified as having a pet face present with a second category-aware classifier trained on a second set of pet images associated with a second category, the processing by the second category-aware cascade classifier identifying zero or more of the potential face location windows as having a pet face present; and verifying the presence of a pet face in each of the potential pet face location windows identified as having a pet face present.