noise removal operation is commonly applied as pre-processing step before subsequent image processing tasks due to the occurrence of noise during acquisition or transmission process. a common problem in imaging systems by using cmos or ccd sensors is appearance of  the salt and pepper noise. this paper presents cellular automata (ca) framework for noise removal of distorted image by the salt an...

We consider binary classification problems with positive definite kernels and square loss, and study the convergence rates of stochastic gradient methods. We show that while the excess testing loss (squared loss) converges slowly to zero as the number of observations (and thus iterations) goes to infinity, the testing error (classification error) converges exponentially fast if low-noise condit...

In this contribution the performance of Reed-Solomon (RS) Block Turbo Coding (BTC) is studied in a 16-QAM modulated OFDM system when operating on a Rayleigh fading channel. BER and PER performances are shown and compared to the performance of Convolutional Turbo Coding (CTC). BTC proves to mitigate the irreducible error floor that limits the performance of CTC at high signal-to-noise ratios (SNR).

Joint Source Channel Coding for Non-Ergodic Channels: The Distortion Signal-to-Noise Ratio (SNR) Exponent Perspective. (May 2008) Kapil Bhattad, B.Tech., Indian Institute of Technology, Madras Chair of Advisory Committee: Dr. Krishna Narayanan We study the problem of communicating a discrete time analog source over a channel such that the resulting distortion is minimized. For ergodic channels,...

Many engineering applications require extracting a signal from observations corrupted by additive noise, possibly heavy-tailed. We assume that the observation noise is a Levy process, while the signal is Gaussian, and derive a non-linear recursive filter that minimizes the L error. A sub-optimal filter is proposed for numerical purposes, and simulations show that it out-performs the existing li...

The two hypotheses are usually distinguished by means of a test-statistic which is an “averaged” quantity computed by combining many observations. When a completely trustworthy ergodic probabilistic model exists for the fading and noise processes, then the probabilities of detection error will drop to zero with the number of observations as the test-statistic concentrates and a threshold can be...