Lossy Compression
I am particularly interested in the use of backward adaptive quantization technique in lossy wavelet image compression. Backward adaptive quantization is meant to be a certain types of adaptive quantization strategies which predict and adapt to varying statistical charateristics of the input signal by using only the causal quantized data as the information source for prediction.

• Context-Based Classification and Adaptive Quantization (CBCAQ) Wavelet Image Coder
  As we operate the compression algorithm in the wavelet transform domain, we consider two layers of adaptation. First, context-based classification of the transform subband data is performed in a backward adaptive manner. That is, given the previously quantized/transmitted wavelet coefficients along the selected scanning path in the wavelet domain, the activity--measured by estimated standard deviation as a function of the quantized past coefficients--of the current coefficient is compared with a set of classification thresholds. This generates multiple wavelet coefficient classes on the fly. Then the second phase of adaptation is achieved within each of the coefficient classes, via parametric adaptation of the Uniform Threshold Quantizer (UTQ), which is employed for subband data quantization. The resulting image coder outperforms most state-of-the-art image coders in the literature, especially at low rates.
  • A new high-performance wavelet image coder based on backward adaptive quantization of subband coefficients. Documentation is available in PS format (gzipped).
  • Superior PSNR numbers and excellent visual quality of reconstructed images
  • Comparison with other state-of-the-art wavelet image coders is available from the UCLA Image Compression Lab
• Related Publications
  • Youngjun Yoo, "Image Subband Coding Using Progressive Classification and Adaptive Quantization", In Proc. of the 4th Human Tech Thesis Prize, Korea, 1998
  • Youngjun Yoo, Antonio Ortega, and Bin Yu, "Image Subband Coding Using Context-Based Classification and Adaptive Quantization," Submitted to IEEE Trans. on Image Proc., Jun. 1997
  • Youngjun Yoo, Antonio Ortega, and Bin Yu, "Adaptive Quantization of Image Subbands with Efficient Overhead Rate Selection," in Proc. ICIP'96, Sep. 16-19, Lausanne, Switzerland
    

Lossless Compression
Wavelet tranform coding methods provide excellent performance for lossy compression of natural images over all compression rates. However wavelet-based mathods fall short when the histogram of the input image, or its subimage, shows only a small number of active intensity levels, i.e., a lot of intensity levels are never used to represent the image/subimage. We call such images simple. The examples of simple image of our interest include: bi-level images; bi-level images scanned in gray scale; computer generated graphics with simple texture, etc.
    Since wavelet transform is not very effective in compressing simple images, image domain compression algorithm can be employed alternatively to achieve good compression. A similar approach can be found in the development and improvement of CREW which opts for bitplane coding using JBIG, an image domain algorithm, with an appropriate criteria to determine simple image/subimages. Equipped with image domain compression option, CREW achieved a fundamental gain in compression performance for simple images and compound images containing simple subimages.

• The Embedded Image-Domain Adaptive Compression (EIDAC) Algorithm for Compression of Simple Images
  The key idea for our proposed image domain compression algorithm is bit plane coding with an adaptive binary arithmetic coder which makes use of probability estimation conditioned on the previously encoded adjacent bits from the upper bit planes as well as within the same bit plane.
      The performance of EIDAC is excellent so that it is adopted as a part of the JPEG2000 standard proposal by USC and TI. Also a related patent application is in the process.
      A future research issue in applying EIDAC is to use in conjunction with wavelet-based codecs in order to compress compound images--which has both simple and natural regions.
• Related Publications
  • Youngjun Yoo, Young Gap Kwon, and Antonio Ortega, "Embedded Image-Domain Adaptive Compression of Simple Images," in Conf. Record of the 32nd Asilomar Conf. on Signals, Sys, and Computers, Nov. 1998
    
  • Youngjun Yoo, Christos Chrysafis, Antonio Ortega, and Jie Liang, Provisional Patent Application related to TI/USC joint JPEG 2000 proposal, Oct. 1997
    
  • Jie Liang, Christos Chrysafis, Antonio Ortega, Youngjun Yoo, Kannan Ramchandran, and Xuguang Yang, "The Predictive Embedded Zerotree Wavelet (PEZW) Coder: A Highly Scalable Image Coder for Multimedia Applications," Proposal for JPEG 2000, Sep. 1997
    

Joint JPEG 2000 Proposal by University of Southern California and Texas Instruments
In the joint project of USC and Texas Instruments for a JPEG2000 proposal, I exclusively worked on the development of an algorithm which aims at lossless compression of simple images.