Home | Research | Publications | People | School | University | Download | Join Us

Examples of Our Work

Interactive visual information processing, interactive image retrieval, segmentation .... (link ...)

Example-based Image Processing  ...

Unifying histogram-based image content descriptor for content-based image retrieval - the CPAM and AI framework  ...  (link ...)

Integrated framework for image coding, indexing, and content-based image retrieval ... (link ...)

Visual guided browsing and navigation for fast image retrieval ... (link ...)

Fast comprehensive tone mapping for high dynamic range image visualization ... (link ...)

Colorizing black and white photos ... (link ....)

Making beautiful pictures,  computational photography ... (link ....)

Machine learning, pattern recognition ... (link ...)

Image data mining, image and feature co-clustering .... (link ...)

Schematics of our example-based image super-resolution system

Example-based Image Processing

We are one of the first groups to propose example-based or learning based image processing technologies. We have developed several lines of application in this area from image super-resolution and coding artifact filtering/removal to the display of high dynamic range image.


Learning-based Image Super-resolution

  1. G. Qiu, "A progressively predictive image pyramid for efficient lossless coding", IEEE Trans. on Image Processing, Vol. 8, No.1, pp. 109 – 115, 1999 (PDF)
  2. G. Qiu, "Inter-resolution look-up table for improved spatial magnification of images", Journal of Visual Communications and Image Representation, vol. 11, pp. 360 - 373, Academic Press, November 2000 (PDF)



Our technique first learns a model that takes a lower resolution version of an image as input and predict a higher resolution version of the same image as output.



Once the model is learned, it is used to predict a higher resolution output from a lower resolution input.



In order to reduce the burden on the learner, we use a smooth prior, i.e., we learn the Laplacian image and add it back to the Gaussian image.

For more detail, please refer to the publications listed above. Work by others that learns to enhance image resolution include the following:

References
  1. W.T. Freeman and E.C. Pasztor, “Learning to Estimate Scenes from Images,” Adv. Neural Information Processing Systems, M.S. Kearns, S.A. Solla, and D.A. Cohn, eds., vol. 11, MIT Press, Cambridge, Mass., 1999, pp. 775-781
  2. W.T. Freeman, E.C. Pasztor, and O.T. Carmichael, “Learning Low-Level Vision,” Int’l J. Computer Vision, vol. 40, no.1, Oct. 2000, pp. 25-47
  3. W.T. Freeman and E.C. Pasztor, “Markov Networks for Superresolution,” Proc. 34th Ann. Conf. Information Sciences and Systems (CISS 2000), Dept. Electrical Eng., Princeton Univ., 2000
  4. W. T. Freeman, T. R. Jones and E. C Pasztor, “Example-Based Super-Resolution”, IEEE Computer Graphics and Applications, Vol. 22, pp. 56-65, March 2002
  5. S. Baker and T. Kanade, “Limits on Super-Resolution and How to Break Them”, IEEE Trans PAMI (24), pp. 1167-1183, September 2002, Earlier: CVPR00 (II: 372-379)

Learning-based Approach to Image Coding Artifact Removal



G. Qiu, "MLP for adaptive postprocessing block coded images", IEEE Transactions on Circuits and Systems for Video Technology, vol. 10, pp. 1450 - 1454, Decemner, 2000  (PDF)

We have developed a method based on the concept of learning-by-examples for blocking-artifact removal in block-coded images. In the developed scheme, inter-block slopes of the compressed image are used as input, the difference between the original uncompressed and the compressed image is used as desired output for training neural networks. Blocking-artifact removal is realized by adding the neural network’s outputs to the compressed image. The new technique has been applied to process JPEG compressed images. Experimental results show significant improvements in both visual quality and peak signal-to-noise ratio. It is also shown the present method is comparable to other state of the art techniques for quality enhancement in block-coded images.


Learning to display high dynamic range images



J. Duan, G. Qiu and G. D. Finlayson,"Learning to display high dynamic range images", CGIV'2004, IS&T's Second European Conference on Color in Graphics, Imaging and Vision,  Aachen, Germany, April 5-8, 2004 (PDF)

We have developed a learning based method to map high dynamic range scenes to low dynamic range images for visualization. We formulate the problem as a quantization process and employ an adaptive learning strategy to ensure that the low dynamic range displays not only faithfully reproduce the original scenes but also are visually pleasing. This is achieved by the use of a competitive learning neural network that employs a frequency sensitive competitive learning mechanism.
Copyright © 2005 G Qiu, Last modified September 2005