Showing papers by "Nalini K. Ratha published in 1997"

System and method for distortion control in live-scan inkless fingerprint images

Abstract: A computer system and method determines the force and/or torque applied during the image acquisition stage of a biometric characteristic. Images with very high or very low pressure or high shear torque are rejected and user/operator is notified to re-acquire the image. Alternatively, the application of force and torque by the subject is restricted mechanically so that the images are acquired while the force and/or torque are within acceptable ranges.

FPGA-based computing in computer vision

Abstract: Algorithms in computer vision are characterized by (i) complex and repetitive operations; (ii) large amount of data and (iii) a variety of data interaction (e.g., point operations, neighborhood operations, global operations). Based on the computation and communication complexity, vision algorithms have been characterized into three categories: (i) low-level, (ii) intermediate-level and (iii) high-level. In this paper, we describe the usage of custom computing approach to meet the computation and communication needs of computer vision algorithms. By customizing hardware architecture for every application at the instruction level, the optimal grain size needed for the problem at hand and the instruction granularity can be matched. Field Programmable Gate Array (FPGA) based processing elements (PEs) are being used to provide this facility. Using programmable communication resources, the diverse communication requirements can be met. A vision system needs to integrate hardware for the three levels. A custom computing approach alleviates the problem of achieving optimal granularity for different stages as the same hardware gets reconfigured at a software level for different levels of the application. We demonstrate the advantages of our approach using Splash 2-a Xilinx 4010-based custom computer.

Multilayer perceptrons on Splash 2

Abstract: Multilayer perceptrons (MLPs) are one of the most popular neural network models for solving pattern classification and image classification problems. Because of their ability to learn complex decision boundaries, MLPs are used in many practical computer vision applications involving classification (or supervised segmentation). Once the connection weights in a MLP have been learnt, the network can be used repeatedly for classification of new input patterns. Several special-purpose architectures have been described in the literature for neural networks as they are slow on a conventional uniprocessor. In this paper, we describe mapping of MLPs onto Splash 2-a "custom computing machine". The main features of the proposed mapping are: (i) the number of nodes in a layer is not fixed; (ii) the number of layers in the network is not fixed; (iii) it is based on a set of reprogrammable FPGAs and a programmable crossbar; and (iv) it has a significant speedup over a uniprocessor. The mapping has been used for implementing a 3-layer MLP for page segmentation application with an appreciable speedup of approximately 150 over a SPARCstation 20 for one million pattern vectors with 20 features per pattern.