We investigate the capability of low level feature detectors to consistently define feature keypoints in an image and its horizontally reflected (mirrored) image. It is our assertion that this consistency is a useful attribute of a feature detector and should be considered in assessing the robustness of a feature detector. We test ten of the most popular detectors using a popular dataset of 8,677 images. We define a set of error measurements to help us to understand the invariance in keypoint position, size and angle of orientation, and we use SIFT descriptors extracted from the keypoints to measure the consistency of extracted feature descriptors. We conclude that the FAST and STAR detectors are perfectly invariant to bilateral symmetry, Good Features to Track and the Harris Corner detector produce consistent keypoints that can be matched using feature descriptors, and others vary in their invariance. SIFT is the least invariant of all the detectors that we test.

In this position paper, we consider the state of computer vision research with respect to invariance to the horizontal orientation of an image – what we term reflection invariance. We describe why we consider reflection invariance to be an important property and provide evidence where the absence of this invariance produces surprising inconsistencies in state-of-the-art systems. We demonstrate inconsistencies in methods of object detection and scene classification when they are presented with images and the horizontal mirror of those images. Finally, we examine where some of the invariances are exhibited in feature detection and descriptors, and make a case for future consideration of reflection invariance as a measure of quality in computer vision algorithms.

Hough Forests have demonstrated effective performance in object detection tasks, which has potential to translate to exciting opportunities in pattern search. However, current systems are incompatible with the scalability and performance requirements of an interactive visual search. In this paper, we pursue this potential by rethinking the methods of Hough Forest training and regression to devise a system that is synonymous with a search index that can yield pattern search results in near real-time. The system performs well on simple pattern detection, demonstrating the concept is sound. However, detection of patterns in complex and crowded street-scenes is more challenging. Some success is demonstrated in such videos, and describe future work that will address some of the key questions arising from our work to date.

Closed-circuit television cameras are used extensively to monitor streets for the security of the public. Whether passively recording day-to-day life, or actively monitoring a developing situation such as public disorder, the videos recorded have proven invaluable to police forces world wide to trace suspects and victims alike. The volume of video produced from the array of camera covering even a small area is large, and growing in modern society, and post-event analysis of collected video is a time consuming problem for police forces that is increasing. Automated computer vision analysis is desirable, but current systems are unable to reliably process videos from CCTV cameras. The video quality is low, and computer vision algorithms are unable to perform sufficiently to achieve usable results. In this chapter, we describe some of the reasons for the failure of contemporary algorithms and focus on the fundamental task of feature correspondence between frames of video - a well-studied and often considered solved problem in high quality videos, but still a challenge in low quality imagery. We present solutions to some of the problems that we acknowledge, and provide a comprehensive analysis where we demonstrate feature matching using a 138-dimensional descriptor that improves the matching performance of a state-of-the-art 384-dimension colour descriptor with just 36% of the storage requirements.

We assess configuration parameters of a Hough Forest and their impact on performance to achieve good pattern detection in images without need or opportunity for large-scale training. Our motivation is the application of Query-by-Example video search where a region of interest in a video frame image is selected by the user as a query sample. We describe a method to discover training images from the single query region and assess the impact of tuning standard Hough Forest parameters on the application of pattern detection using a very small training set of data. We introduce a novel method by which to improve training performance and precision in pattern detection by adaptive selection of the patch size and the number of patches used. We show results using challenging street-scene videos demonstrating our method to improve precision and performance over static general-purpose Hough Forest parameters used in the literature.

We describe a methodology that is designed to match key point and region-based features in real-world images, acquired from long-running security cameras with no control over the environment. We detect frame duplication and images from static scenes that have no activity to prevent processing saliently identical images, and describe a novel blur sensitive feature detection method, a combinatorial feature descriptor and a distance calculation that efficiently unites texture and colour attributes to discriminate feature correspondence in low quality images. Our methods are tested by performing key point matching on real-world security images such as outdoor CCTV videos that are low quality and acquired in uncontrolled conditions with visual distortions caused by weather, crowded scenes, emergency lighting or the high angle of the camera mounting. We demonstrate an improvement in accuracy of matching key points between images compared with state-of-the-art feature descriptors. We use key point features from Harris Corners, SIFT, SURF, BRISK and FAST as well as MSER and MSCR region detectors to provide a comprehensive analysis of our generic method. We demonstrate feature matching using a 138-dimensional descriptor that improves the matching performance of a state-of-the-art 384-dimension colour descriptor with just 36% of the storage requirements.

Finding corresponding key points in images from security camera videos is challenging. Images are generally low quality and acquired in uncontrolled conditions with visual distortions caused by weather, crowded scenes, emergency lighting or the high angle of the camera mounting. We describe a methodology to match features between images that performs especially well with real-world images. We introduce a novel \emph{blur sensitive feature detection} method, a combinatorial feature descriptor and a distance calculation that efficiently unites texture and colour attributes to discriminate feature correspondence in low quality images. Our methods are tested by performing key point matching on real-world security images such as outdoor CCTV videos, and we demonstrate an improvement in the ability to match features between images compared with the standard feature descriptors extracted from the same set of feature points. We use key point features from Harris Corners, SIFT, SURF, BRISK and FAST as well as MSER and MSCR region detectors to provide a comprehensive analysis of our generic method. We demonstrate feature matching using a 138-dimensional descriptor that improves the matching performance of a state-of-the-art 384-dimension colour descriptor with just 40% of the storage requirements.

@inproceedings{Henderson2015,
address = {London},
author = {Henderson, Craig and Izquierdo, Ebroul},
booktitle = {Science and Information Conference},
title = {{Robust Feature Matching in the Wild}},
year = {2015}
}