Sunil Kumar Kopparapu

Lighting design for machine vision application

Abstract Low-level image processing is an essential first step in any machine vision application. Low-level vision processing tasks need good lighting in the work environment for them to function robustly. Hence, good and uniform illumination from external light source is essential for machine vision applications to function. In this paper, we suggest a design procedure to obtain uniform illumination on the scene being imaged using several light sources. We pose the problem of determining the optimal position of the light sources as a minimisation problem. Simulation results show the effectiveness and suitability of the proposed procedure to illuminate the scene uniformly.

Choice of Mel filter bank in computing MFCC of a resampled speech

Mel Frequency Cepstral Coefficients (MFCCs) are the most popularly used speech features in many speech and speaker recognition applications. In this paper, we study the effect of resampling a speech signal on these speech features. We first derive a relationship between the MFCC parameters of the resampled speech and the MFCC parameters of the original speech. We propose six methods of calculating the MFCC parameters of downsampled speech by transforming the Mel filter bank used to compute MFCC of the original speech. We then experimentally compute the MFCC parameters of the down sampled speech using the proposed methods and compute the Pearson coefficient between the MFCC parameters of the downsampled speech and that of the original speech to identify the most effective choice of Mel-filter band that enables the computed MFCC of the resampled speech to be as close as possible to the MFCC of the original speech.

SMS based natural language interface to yellow pages directory

Yellow Pages are directories that source information about various commercial organizations like their addresses, phone contact and other details. These are very useful and are used by individual and other business houses. Until recently, the only way to access these yellow pages directory information was to physically look into a huge hard-copy directory, which was not only laborious but also time consuming and required the user to be familiar with the organization of the directory. More recently, there have been IVR based contact centers that have been set up which can be used by the users to query information. While it is easier than browsing through the physical directory, it still has several pitfalls. The time spent on trying to get the information is quite long and at the end of enquiry one is not sure if one will get the information that one is looking for. In this paper, we propose a novel interface which enables accessing the yellow pages directory information on the mobile phone by sending a short message service (SMS). The central idea of the proposed method is to avoid any constraint on the way the user can query the yellow pages directory except that it be in natural English. The system, which uses natural language processing (NLP) techniques, understands the intent of the query and intelligently searches the yellow pages directory to retrieve information. This retrieved information is then sent back to the user in the form of a SMS.

Edge assisted fast binarization scheme for improved vehicle license plate recognition

Identity of a motor vehicle is usually through its license plate. Automatic vehicle license plate recognition has several applications in intelligent traffic management systems. In this paper, we propose and describe a fast edge assisted adaptive binarization technique for improved extraction of text from license plate images captured using mobile phone camera. We evaluate and compare the performance of the proposed binarization scheme with some well known algorithms on real vehicle images. Experiments on 400 real vehicle images captured using mobile phones shows that the edge based scheme is not only fast but also performs as well as the other binarization technique.

Fuzzy Directional Features for unconstrained on-line Devanagari handwriting recognition

This paper describes a novel feature set for recognition of unconstrained on-line handwritten Devanagari script. Experiments are conducted for the automatic recognition of handwritten character primitives (sub-character units) collected without any constraints from different writers. Initially we describe the Fuzzy Directional Feature (FDF) extraction method and then show how these features can be effectively utilized for writer independent Devanagari character recognition. The recognition algorithm uses second order statistics to construct different stroke models. Experimental results show that FDF set out performs commonly used Directional Features (DF) for writer independent data set at stroke level recognition.

A spoof resistant multibiometric system based on the physiological and behavioral characteristics of fingerprint

Despite emerging as a prominent choice to serve the security concerns of person authentication applications, unimodal biometric systems are vulnerable to spoof attacks. Multimodal biometric systems can effectively minimize spoof attacks while improving the overall performance. In this paper, we present a multimodal system based on two modalities derived from multi instance fingerprint acquisition viz. fingerprint and the associated time dynamics. Extensive user verification and spoof resistance experiments conducted on virtual multimodal databases, created by combining ATVS and LivDet-13 fingerprint databases each with fingerprint dynamics database. Fusion is performed at match score level using sum and weighted sum rules. The empirical results demonstrate spoof resistance of the proposed multimodal system with significant performance improvement over unimodal and multi-instance fingerprint recognition systems. The performance of the proposed system is evaluated on well-known metrics like Detection Error Trade-off (DET) curves, equal error rate (EER), and Area Under the Curve (AUC). Display Omitted Multimodal system based on physiological and behavioral characteristics is proposed.Experiments are conducted in scenarios: unimodal, multi-instance and multimodal.The system performance evaluation is performed in the presence of spoof samples.Average relative improvement in EER over unimodal is 90.64%.Average relative improvement in EER over multi-instance system is 82.57%.

Identifying Optimal Gaussian Filter for Gaussian Noise Removal

In this paper we show that the knowledge of noise statistics contaminating a signal can be effectively used to choose an optimal Gaussian filter to eliminate noise. Very specifically, we show that the additive white Gaussian noise (AWGN) contaminating a signal can be filtered best by using a Gaussian filter with specific characteristics. The design of the Gaussian filter bears relationship with the noise statistics in addition to some basic information about the signal. We first derive a relationship between the properties of the Gaussian filter, the noise statistics and the signal and later show through experiments that this relationship can be used effectively to identify the optimal Gaussian filter that can effectively filter noise.

A novel approach to identify problematic call center conversations

Voice based call centers enable customers to query for information by speaking to agents in the call center. Most often these call conversations are recorded for analysis with the intent of trying to identify things that can help improve the performance of the call center to serve the customer better. Today the recorded conversations are analyzed by humans by listening to call conversations, which is both time consuming, fatigue prone and not accurate. Additionally, humans are able to analyze only a small percentage of the total calls because of economics. In this paper, we propose a visual method to identify problem calls quickly. The idea is to sieve through all the calls and identify problem calls, these calls can then be further analyzed by human. We first model call conversations as a directed graph and then identify a structure associated with a normal call. All call conversations that do not have the structure of a normal call are then classified as being abnormal. In this paper, we use the speaking rate feature to model call conversation because it makes it easy to spot potential problem calls. We have experimented on real call center conversations acquired from a call center and the results are encouraging.

A two pass algorithm for speaker change detection

Speaker change detection is a necessary first step in several applications. In this paper, we propose an unsupervised two pass algorithm for speaker change detection in conversational speech. Generalized Likelihood Ratio (GLR) metric is used in the first pass to coarsely identify speaker change points and during the second pass, these candidate change points are finely analyzed assuming that the initial part of the conversational audio is from one of the speakers. The final change point detection decision is based on the likelihood probability function computed for the segments between two consecutive candidate change points using the known speaker model. The proposed two pass algorithm has been tested on a question and answer session of a financial audio report of a company and also on an audio track of a movie.

Multimodal Indexing of Multilingual News Video

The problems associated with automatic analysis of news telecasts are more severe in a country like India, where there are many national and regional language channels, besides English. In this paper, we present a framework for multimodal analysis of multilingual news telecasts, which can be augmented with tools and techniques for specific news analytics tasks. Further, we focus on a set of techniques for automatic indexing of the news stories based on keywords spotted in speech as well as on the visuals of contemporary and domain interest. English keywords are derived from RSS feed and converted to Indian language equivalents for detection in speech and on ticker texts. Restricting the keyword list to a manageable number results in drastic improvement in indexing performance. We present illustrative examples and detailed experimental results to substantiate our claim.