N. Puviarasan

An investigation of neuro-fuzzy systems in psychosomatic disorders

A neuro-fuzzy model for diagnosis of psychosomatic disorders is proposed in this paper. The symptoms and signs are collected from the patients through oral interview. For the linguistic nature of patients inputs, an artificial domain is created and fuzzy membership values are defined. The fuzzy values are fed as inputs to feedforward multilayer neural network. The network is trained using Backpropagation training algorithm. The trained model is tested with new patients symptoms and signs. Further, the performance of the diagnosing capability is compared with medical expert. The performance of the model is also compared with probability model based on Bayesian Belief Network and statistical model using Linear Discriminant analysis

Lip reading of hearing impaired persons using HMM

This paper describes a method for lip reading of hearing impaired persons. The term lip reading refers to recognizing the spoken words using visual speech information such as lip movements. The visual speech video of the hearing impaired person is given as input to the face detection module for detecting the face region. The region of the mouth is determined relative to the face region. The mouth images are used for feature extraction. The features are extracted using discrete cosine transform (DCT) and discrete wavelet transform (DWT). Then, these features are applied separately as inputs to the hidden markov model (HMM) for recognizing the visual speech. To understand the visual speech of hearing impaired person in cash collection counters, 33 words are chosen. For each word, 20 samples are collected for training the HMM model and another five samples are used for testing the model. The experimental results show that the method gives the performance of 91.0% for the DCT based lip features and 97.0% for DWT based lip features.

Diagnosis of gastrointestinal disorders using DIAGNET

Abstract A new neural network model called DIAGNET is proposed in this paper for diagnosing gastrointestinal disorders. DIAGNET is a combination of Backpropagation neural network (BPNN) and radial basis functions neural network (RBFNN). The symptoms and signs are collected from the patients through oral interview. For the linguistic nature of patient’s inputs, an artificial domain is created and fuzzy membership values are defined. The fuzzy values are fed as inputs to the DIAGNET and trained for diagnosing the diseases related to gastrointestinal disorders. The trained model is tested with new patient’s symptoms and signs. The performance of the DIAGNET is compared with the existing Backpropagation neural network and Radial basis functions neural network models. Sensitivity, Specificity and Receiver-Operating Characteristics (ROC) are used as the indicators for testing the accuracy of the models which predict the gastrointestinal disorder diseases. The results suggest that the DIAGNET can be better solution for complex, nonlinear medical decision support systems.

Color Image Compression Based on Feature Extraction

This paper proposes an efficient compression scheme for compressing RGB color images based on feature extraction with the combination of DCT transform and the Peano-Hilbert Scan. The RGB color image is converted into YCbCr in order to extract the color and the texture features. The DCT transform is applied to the extracted luma and the chroma component to reduce the redundancy. Peano-Hilbert scanning is performed over the DCT matrix which increases the PSNR of the reconstructed image. The proposed bi-mode quantization is applied to preserve the image quality. The quantized coefficients are encoded using the lossless Huffman encoding. The efficiency of the proposed compression scheme has been implemented and compared with other existing compression techniques. The proposed compression method can achieve a higher fidelity and faster decompression time compared to other lossy compression methods. Thus, Compression based on feature extraction contributes to better performance.

Tri-mode dual level 3-D image compression over medical MRI images

Digital image sequence requires huge storage and high bandwidth for transmitting data in uncompressed form of the multimedia communication. Efficient image compression technique is required to meet the acceptable quality. In this paper, an efficient image compression technique is developed based on the intensity value of the pixels. The proposed image compression algorithm contains two levels of compression. Initially the average between the neighboring of a particular pixel is evaluated and it is assigned to the pixels of the original image. This reduces the correlations of the intensity level. The same pixel assignment is extended towards three dimensional forms along rows, column and the diagonals. To retain the quality of the image the compressed image based on such intensity assignment is encoded using the proposed tri-mode encoding scheme. The encoded bits are decoded at the compression decoder. To reveal the efficiency of the proposed method, the obtained results are compared with the other existing methods. The proposed algorithm has been tested in magnetic resonance imaging (MRI) images of the brain. Simulation results show the proposed method gives the best results over the other existing methods.