Some Other Projects

Learning with progressive transductive support vector machine (2002)

Keywords

Pattern classification, Statistical learning, Support vector machines, Transductive inference, Semi-supervised learning.

Brief Description

In this approach, a progressive transductive support vector machine (PTSVM) is proposed to extend classical transductive support vector machine (SVM) to handle different class distributions. It solve the problem of having to estimate positive/negative ratio from the working set. Following is a simple illustration.

 

 

There are altogether 6 labeled examples. The three "+" points are positive and the three "o" points are negative. The solid line the the final separating hyperplane found by PTSVM and the dashed line the the final hyperplane found by TSVM. PTSMV avoids the problem of unreliable ratio estimation by a progressive training strategy and results in better classification precission.   

 

 

Coarse scale cost-utility approach for efficient unequal loss protection (2002)

Keywords

Scalable image/video coding, Error resilience, Forward error correction, Unequal loss protection, Wavelet coding.

Brief Description

In this approach, an efficient preprocessing algorithm is presented based on the idea of the coarse scale cost-utility working curve. This method significantly improves the efficiency of the multiple description based unequal loss protection framework with substantial memory and computation saving. The idea of coarse scale operation can also be successfully employed in many other applications to improve performances.

Fast bit allocation in sub-band coding (2001)

Keywords

Wavelet image coding, Optimal bit allocation, Constrained optimization, Lagrange multiplier method, Dynamic programming.

Brief Description

In this approach, a simple but efficient algorithm is presented to speed up the convergence of the Lagrange multiplier method based bit allocation in the context  of wavelet image coding based on the concave property of the envelope of the multiplier line cluster. Experimental result shows that our new algorithm is about an order of magnitude faster than the original algorithm.

 

Feature difference classification method in fractal image coding (2000)

Keywords

Fractal image coding, Iterated function system, Contraction mapping principle, Domain-range matching, Clustering, Kd-tree.

Brief Description

In this approach, a classification algorithm is presented to rule out pseudo domain-range matches effectively. The algorithm based on the contractive characteristics of transformations in fractal image coding. It can lead to significant improvement of the rate-distortion performance.

 

Fast fine granularity decoding of fractal image coding (2000)

Keywords

Fractal image compression, Iterated function system, Scalable codec, Resolution/Quality scalability, Gauss-Seidel Iteration.

Brief Description

In this approach, a single-buffer based fine granularity iterative decoding is proposed to achieve faster convergence. Following is an intuitive comparison of several  reconstructed images after 3 iterations under different decoding strategies. The superiority of our algorithm over other algorithms is obvious.

(a) The original Goldhill image    (b) Conventional decoding(PSNR=27.06dB)    (c) Pixel update(PSNR=30.82dB)    (d) Ordered pixel update(PSNR=31.74dB) 

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~