Mantosh Biswas

A generalized image denoising method using neighbouring wavelet coefficients

The NeighShrink and ModiNeighShrink areefficient image denoising algorithms that are based on universal threshold and discrete wavelet transform. The improved image denoising method based on wavelet thresholding (IIDMWT) method gives better results than the NeighShrink and ModiNeighShrink by using the modified universal threshold. These methods kill too many wavelet coefficients; some of them may contain useful image information. Thus, we may not get good quality of image using these methods. In this paper, we extend the idea of Cai and Silverman for developing a new image denoising method and determine the coefficients of neighboring window for every subband. The experimental results show that in most of the cases, our proposed method performs better than the NeighShrink, ModiNeighShrink, and IIDMWT in terms of peak signal- to-noise ratio and structural similarity index measure.

Discriminative Training Using Noise Robust Integrated Features and Refined HMM Modeling

Abstract The classical approach to build an automatic speech recognition (ASR) system uses different feature extraction methods at the front end and various parameter classification techniques at the back end. The Mel-frequency cepstral coefficients (MFCC) and perceptual linear prediction (PLP) techniques are the conventional approaches used for many years for feature extraction, and the hidden Markov model (HMM) has been the most obvious selection for feature classification. However, the performance of MFCC-HMM and PLP-HMM-based ASR system degrades in real-time environments. The proposed work discusses the implementation of discriminatively trained Hindi ASR system using noise robust integrated features and refined HMM model. It sequentially combines MFCC with PLP and MFCC with gammatone-frequency cepstral coefficient (GFCC) to obtain MF-PLP and MF-GFCC integrated feature vectors, respectively. The HMM parameters are refined using genetic algorithm (GA) and particle swarm optimization (PSO). Discriminative training of acoustic model using maximum mutual information (MMI) and minimum phone error (MPE) is preformed to enhance the accuracy of the proposed system. The results show that discriminative training using MPE with MF-GFCC integrated feature vector and PSO-HMM parameter refinement gives significantly better results than the other implemented techniques.

Discriminative Training using Heterogeneous Feature Vector for Hindi Automatic Speech Recognition System

Training and testing are the two phases that are used in statistical approach of designing an automatic speech recognition (ASR) system. The training phase includes parameterization of input speech signal and acoustic modeling of speech features. The paper proposes discriminative training of hidden Markov Model (HMM) that uses heterogeneous feature vector for continuous Hindi ASR system. A linear interpolation of mel frequency cepstral coefficients (MFCC) and perceptual linear prediction (PLP) is used to generate heterogeneous feature streams. The implemented work uses maximum mutual information estimation (MMIE) and minimum phone error (MPE) discriminative techniques for acoustic model training. The results show that MF-PLP parameterization with MPE discriminative techniques combination outperforms the other feature extraction and discriminative combination.

Adaptive histogram equalization based fusion technique for hazy underwater image enhancement

Degradation of underwater images is an atmospheric phenomenon which is a result of scattering and absorption of light. In this paper, we have defined a fusion based approach to enhance the visibility of underwater images. Our method uses only one single hazy image to derive the contrast improved and colour corrected versions of the original image. Further, it removes the distortion and uplifts the visibility of the distant objects in the image by applying weight maps on each of the derived inputs. We have used multi-scale fusion technique to blend the inputs and weight maps together, ensuring that each fused image contributes its most significant feature into the final image. Our technique is simple and straightforward that effectively contribute in enhancing the quality and appearance of underwater hazy images.

An adaptive wavelet thresholding image denoising method

The NeighShrink, IAWDMBNC, and IIDMWT are important methods to remove the noise from a corrupted image. These methods however cannot recover original image significantly since the threshold value does not minimize the noisy wavelet coefficients across scales and thus they do not give good quality of image. In this paper, we propose an adaptive denoising method that provides an adaptive way of setting up minimum threshold by shrinking the wavelet coefficients to overcome the above problem using an exponential function. Our method retains the original image information efficiently by removing noise and it has the image quality parameters such as peak-to-signal nose ratio (PSNR) and Structural Similarity Index Measure (SSIM) better than the NeighShrink, IAWDMBNC, and IIDMWT methods.

Selective Parameters Based Image Denoising Method

In this paper, we propose a Selective Parameters based Image Denoising method that uses a shrinkage parameter for each coefficient in the subband at the corresponding decomposition level. Image decomposition is done using the wavelet transform. VisuShrink, SureShrink, and BayesShrink define good thresholds for removing the noise from an image. SureShrink and BayesShrink denoising methods depend on subband to evaluate the threshold value whereas the VisuShrink is a global thresholding method. These methods remove too many coefficients and do not provide good visual quality of the image. Our proposed method not only keeps more noiseless coefficients but also modifies the noisy coefficients using the threshold value. We experimentally show that our method provides better performance in terms of objective and subjective criteria i.e. visual quality of image than the VisuShrink, SureShrink, and BayesShrink.

Edge Detection Technique Using ACO with PSO for Noisy Image

In image processing, the edges of an image are those pixels whose intensity values changes drastically. Various techniques have been applied which rely on the ant colony optimization (ACO) for edge detection and the threshold value calculated for edge detection technique is either user defined or taken as the mean value of the obtained pheromone matrix. Other challenges to deal with edge detection are the presence of noisy environment, so to deal with it, in this paper, we define adaptive threshold value based on particle swarm optimization (PSO) for edge detection to overcome the limitation of existing ACO-based edge detection techniques. The experiment results have shown that the proposed technique has performed better under noisy environment over Sobel, Canny, and ACO-based technique both for objective criteria, i.e., restored edge images and subjective criteria, i.e., PSNR, precision, recall, and F-measure for test images in addition to reducing the execution time.

Threshold-Based Clustering of SAR Image Using Gaussian Kernel and Mean-Shift Methods

Image clustering is very useful for recognizing the inner architect of the image data set. It is the process of partitioning the image into clusters such that they have relatively high similarity among the data points within a cluster and have relatively low similarity or no similarity between any two different clusters moreover cluster can help to classify the high-resolution remote sensing data (WorlView2). In this paper, we present a clustering for synthetic aperture radar (SAR) image by threshold-based Gaussian kernel and mean-shift methods. It is a nonparametric process that can detect irregular shape in the spatial and feature range of a SAR images. The main concept of the proposed clustering algorithm is an iterative movement of the data points to their highest density point by calculating the gradient of kernel density estimate. The accuracy and efficiency of mean-shift-based clustering of SAR image are highly reliant on the bandwidth range of the kernel that is user defined. Therefore in this paper, we improved the accuracy and efficiency with by calculating the bandwidth range using statistical techniques on the co-occurrence matrix of SAR image. The proposed method is stable, and it improves the speed and at the same time scale down the over portioning issue.

Edge Detection Algorithm Using Dynamic Fuzzy Interface System

Edge in an image depends on the viewer’s perspective i.e., some viewers may feel it as edge and some may not. Fuzzy logic can be used to solve this partial truth value concept. Many fuzzy-based edge detection methods have been proposed till now, but most of them used the static fuzzy inference system for edge detection, in which we have to change the membership functions for each image in order to get better results. Therefore, to overcome this drawback, we proposed fuzzy logic-based edge detection algorithm with dynamic generation of fuzzy interface system (FIS). The performance of the proposed method is demonstrated through computer simulation results over Sobel, Canny, EFLEDG, and EDFLM edge detection methods in terms of both subjective and fidelity criteria and our proposed method gave good results in terms of F-Measure and visual quality of resultant edge images.

Discriminatively trained continuous Hindi speech recognition system using interpolated recurrent neural network language modeling

Abstract This paper implements and evaluates the performance of a discriminatively trained continuous Hindi language speech recognition system. The system uses maximum mutual information and minimum phone error discriminative techniques with various numbers of Gaussian mixtures to train the automatic speech recognition (ASR) system. The training dataset consists of Hindi speech transcription. The experiments show a significant performance gain over maximum likelihood-based Hindi language speech recognition system. The system uses an efficient recurrent neural network (RNN)-based language modeling. The results indicate that the use of RNN-based language modeling enhances the performance of the ASR system. Further, the interpolation of n-gram language model (LM) with the RNNLM exhibits an additional increase in recognition performance of the implemented system. The proposed system introduces the concept of speaker adaption using maximum likelihood linear regression technique. The paper also gives an overview of the techniques used for discriminative training along with practical issues involved in their implementation.