Gray Scale and Rotation Invariant Texture Classification with Local Binary Patterns
ECCV '00 Proceedings of the 6th European Conference on Computer Vision-Part I
ACM SIGGRAPH 2004 Papers
LOCUS: Learning Object Classes with Unsupervised Segmentation
ICCV '05 Proceedings of the Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1 - Volume 01
Cosegmentation of Image Pairs by Histogram Matching - Incorporating a Global Constraint into MRFs
CVPR '06 Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Volume 1
Image segmentation evaluation: A survey of unsupervised methods
Computer Vision and Image Understanding
The Frankencamera: an experimental platform for computational photography
ACM SIGGRAPH 2010 papers
Hi-index | 0.00 |
The heavy use of camera phones and other mobile devices all over the world has produced a market for mobile image analysis, including image segmentation to separate out objects of interest. Automatic image segmentation algorithms, when employed by many different users for multiple applications, cannot guarantee high quality results. Yet interactive algorithms require human effort that may become quite tedious. To reduce human effort and achieve better results, it is worthwhile to know in advance which images are difficult to segment and may require further user interaction or alternate processing. For this purpose, we introduce a new research problem: how to estimate the image segmentation difficulty level without actually performing image segmentation. We propose to formulate it as an estimation problem, and we develop a linear regression model using image features to predict segmentation difficulty level. Different image features, including graytone, color, gradient, and texture features are tested as the predictive variables, and the segmentation algorithm performance measure is the response variable. We use the benchmark images of the Berkeley segmentation dataset with corresponding F-measures to fit, test, and choose the optimal model. Additional standard image datasets are used to further verify the model's applicability to a variety of images. A new feature that combines information from the log histogram of log gradient and the local binary pattern histogram is a good predictor and provides the best balance of predictive performance and model complexity.