Download presentation
Presentation is loading. Please wait.
Published byCecilia Johnston Modified over 8 years ago
1
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. A framework for improved pedestrian detection performance through blind image distortion categorization. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
2
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. One row of the covariance matrices of neighbor pixels (with DC coefficient removed) of pristine, AWGN corrupted and JPEG compressed images from our dataset. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
3
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. Scatter plots of the grayscale values of neighboring pixels are shown for pristine (a), JPEG compressed (b), and AWGN corrupted (c) images. The values have been scaled so that the mean is zero and variance one. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
4
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. Sample images from our test dataset. A reference image 4 is shown in (a). “Blocking artifacts” caused by JPEG compression are evident on the side of the blue van in (b). AWGN, which can be introduced during image capture, also degrades the quality of the reference image, as illustrated in (c). Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
5
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. (a) Performance of a classifier trained with poor quality AWGN images on all noise corrupted images in the INRIA 4 distorted dataset. (b) Performance of a classifier trained with poor quality JPEG images on all compressed images in the INRIA 4 distorted dataset. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
6
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. (a): LAMRs of pedestrian detection algorithm performance on the AWGN degraded INRIA 4 database. (b) LAMRs of the JPEG compressed INRIA 4 database. It is observed that for a particular range of quality parameter, a small increase in image quality can yield a large increase in detection performance. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
7
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. The performance of our proposed multi-classifier “quality-aware” detection framework offers significant improvement over single classifier detection models on the INRIA 4 dataset. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
8
Date of download: 6/3/2016 Copyright © 2016 SPIE. All rights reserved. The proposed multi-classifier “quality-aware” detection framework is the best performing algorithm (LAMR) on real distortions from the ChangeDetection 55 dataset. Figure Legend: From: Performance optimization for pedestrian detection on degraded video using natural scene statistics J. Electron. Imaging. 2014;23(6):061114. doi:10.1117/1.JEI.23.6.061114
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.