Search

Describable visual attributes are a powerful way to label aspects of an image, and taken together, build a detailed representation of a scene’s appearance. Attributes enable highly accurate approaches to a variety of tasks, including object recognition, face recognition and image retrieval. An important consideration not previously addressed in the literature is the reliability of attribute classiers as the quality of an image degrades. In this paper, we introduce a general framework for conducting reliability studies that assesses attribute classier accuracy as a function of image degradation. This framework allows us to bound, in a probabilistic manner, the input imagery that is deemed acceptable for consideration by the attribute system { without requiring ground truth attribute labels. We introduce a novel dierential probabilistic model for accuracy assessment that leverages a strong normalization procedure based on the statistical extreme value theory. To demonstrate the utility of our framework, we present an extensive case study using 64 unique facial attributes, computed on data derived from the Labeled Faces in the Wild (LFW) data set. We also show that such reliability studies can result in signicant compression benets for mobile applications.

The issues of applying facial recognition at significant distances are non-trivial and often subtle. This paper summarizes 7 years of effort on Face at a distance, which for us is far more than a fad. Our effort started under the DARPA Human Identification at a Distance (HID) program. Of all the programmers under HID, only a few of the efforts demonstrated face recognition at greater than 25ft and only one, lead by Dr. Boult, studied face recognition at distances greater than 50 meters. Two issues were explicitly studied. The first was atmospherics/weather, which can have a measurable impact at these distances. The second area was sensor issues including resolution, field-of-view and dynamic range. This paper starts with a discussion and some of results in sensors related issues including resolution, FOV, dynamic range and lighting normalization. It then discusses the "Photohead" technique developed to analyze the impact of weather/imaging and atmospherics at medium distances. The paper presents experimental results showing the limitations of existing systems at significant distance and under non-ideal weather conditions and presents some reasons for the weak performance. It ends with a discussion of our FASSTTM (failure prediction from similarity surface theory) and RandomEyesTM approaches, combined into the FIINDERTM system and how they improved FAAD.