This paper discusses the potential benefits of applicationspecific power management through remote task execution. Power management is crucial for mobile devices that have to rely on battery power for extended periods of time. Image processing and understanding is a class of applications that is important in mobile environments. Image processing can be used in autonomous robot navigation, target acquisition/classification, keyboard-less input, and aerial surveillance (Micro Air Vehicles), just to mention a few. Experimental results on an image processing application, namely a human face detection and recognition system, indicate the power savings that can be achieved for this important class of applications. We discuss a compilation strategy that generates two versions of the initial application, one to be executed on the mobile device (client), and the other on a machine connected to the mobile device via a wireless network (server). The client and server codes have to be able to deal with disconnection events. Our proposed compilation strategy uses checkpointing techniques to allow the client to monitor program progress on the server, and to request checkpoint data in order to reduce the performance penalty in case of a possible server and/or network failure. The reported results have been obtained by actual power measurements on three client systems, (1) the This research was partially conducted while the first author was a visiting researcher at the Compaq's Cambridge Research Lab (CRL). Additional funding has been provided by NSF CAREER award No. 9985050. StrongARM based low-power SKIFF system developed at Compaq's Cambridge Research Laboratory, (2) Compaq's commercially available StrongARM based iPAQ H3600, and (3) a PentiumII based laptop. Initial experimen...

This paper discusses the potential benefits of applicationspecific power management through remote task execution. Power management is of particular importance for mobile devices that have to rely on battery power for extended periods of time. Image processing is a class of applications important in mobile environments. Image processing can be used in autonomousrobot navigation, target acquisition/classification, keyboard-less input, and aerial surveillance (Micro Air Vehicles), just to mention a few. Experimental results on an image processing application, namely a human face detection and recognition system, indicate the power savings that can be achieved for this important class of applications. We discuss a compilation strategy that generates two versions of the initial application, one to be executed on the mobile device (client), and the other on a machine connected to the mobile device via a wireless network (server). The client and server codes have to be able to deal with disconnectionevents. Our proposed compilation strategy uses checkpointing techniques to allow the client to monitor program progress on the server, and to request checkpoint data in order to reduce the performance penalty in case of a possible server and/or network failure. The reported results have been obtained by actual power measurements on the StrongARM based lowpower SKIFF system developed at Compaq’s Cambridge Research Laboratory. Initial experiments show that energy consumption can be reduced significantly, in some cases up to one order of magnitude, depending on the selected characteristics of the mobile device, remote host, and wireless network.

Abstract—In this paper, we propose an efficient face recognition scheme which has two features: 1) representation of face images by two-dimensional (2-D) wavelet subband coefficients and 2) recognition by a modular, personalised classification method based on kernel associative memory models. Compared to PCA projections and low resolution “thumb-nail ” image representations, wavelet subband coefficients can efficiently capture substantial facial features while keeping computational complexity low. As there are usually very limited samples, we constructed an associative memory (AM) model for each person and proposed to improve the performance of AM models by kernel methods. Specifically, we first applied kernel transforms to each possible training pair of faces sample and then mapped the high-dimensional feature space back to input space. Our scheme using modular autoassociative memory for face recognition is inspired by the same motivation as using autoencoders for optical character recognition (OCR), for which the advantages has been proven. By associative memory, all the prototypical faces of one particular person are used to reconstruct themselves and the reconstruction error for a probe face image is used to decide if the probe face is from the corresponding person. We carried out extensive experiments on three standard face recognition datasets, the FERET data, the XM2VTS data, and the ORL data. Detailed comparisons with earlier published results are provided and our proposed scheme offers better recognition accuracy on all of the face datasets. Index Terms—Face recognition, wavelet transform, associative memory, kernel methods. I.

This paper analyzes the merits of two biologically-inspired face recognition models, eigenfaces and graphmatching, in the context of related neurophysiological and psychophysical data. Given the ambiguity of current biological evidence, a more promising direction for future face recognition research is in the development of models that conform more closely to human perception of facial similarity. c fl2000 Carnegie Mellon University Introduction For both neuroscientists and computer scientists, face recognition is a fascinating problem with important commercial applications such as mug shot matching, crowd surveillance, and witness face reconstruction. Physiological evidence indicates that the brain possesses specialized face recognition hardware in the form of face detector cells in the inferotemporal cortex and regions in the frontal right hemisphere; impairment in these areas leads to a syndrome known as prosapagnosia. Interestingly, prosapagnosics, although unable to recognize f...

Psychophysical studies have shown that humans actively exploit temporal information such as contiguity of images in object recognition. We have recently developed a recognition system which uses temporal contiguity to learn extensible representations of objects on-line. The system performs well both on real-world and synthetic data and shows robustness under illumination changes. In this paper, we present results which compare the proposed representation against simple image-based representations of the same complexity using Minkowski Minimum Distance classifiers and Support Vector Machine classifiers. Recognition results for all classifiers show large improvements with incorporated temporal information. 1

This paper describes an algorithm that uses multi-scale Gaussian differential features (MGDFs) for face recognition. Results on standard sets indicate at least 96 % recognition accuracy, and a comparable or better performance with other well known techniques. The MGDF based technique is very general; its original application included similarity retrieval in textures, trademarks, binary shapes and heterogeneous gray-level collections. 1

ARGUS is a multi-agent visitor identication system distributed over several workstations. Human faces are extracted from security camera images by a neuralnetwork -based face detector, and identied as frequent visitors by ARENA, a memory-based face recognition system. ARGUS then uses a messaging system to notify hosts that their guests have arrived. An interface agent enables users to submit feedback, which is immediately incorporated by ARENA to improve its face recognition performance. The ARGUS components were rapidly developed using JGram, an agent framework that is also detailed in this paper. JGram automatically converts high-level agent specications into Java source code, and assembles complex tasks by composing individual agent services into a JGram pipeline. ARGUS has been operating successfully in an outdoor environment for several months. Introduction Consider the following scenario. Visitors to large apartment complexes are typically screened by a security guard in the ...

We discuss a new nonlinear feature extraction algorithm, the solution of which can be obtained in closed-form. The features can be used for general pattern recognition applications including those involving class representation, class discrimination, or for both simultaneously. The feature extraction method is called the maximum representation and discrimination feature (MRDF) method. A new discrimination measure is used that can handle classes with multiple clusters in the input space. A computationally efficient nonlinear MRDF procedure to extract nonlinear features from high-dimensional input images is devised. This nonlinear feature extraction procedure is shown to provide very general nonlinear surfaces, in contrast with other nonlinear techniques. The nonlinear MRDF is also shown to generalize to well-known linear and nonlinear image processing operations. We show the use of the MRDF feature extraction technique for classification of product inspection items, classification and pose estimation of objects (machined-parts), and pose-invariant face recognition. Comparisons of the results obtained with our MRDF features with other well-known feature extraction procedures such as the Karhunen-Loeve (KL) transform, and the Fisher linear discriminant are also made.

Abstract not found

ppeared in a given subset, and is therefore a random variable (different for each technique), and any given trial of subsampling or k-fold cross-validation is equivalent to a single observation. The best estimate of the generalization accuracy is given by the expectation of this random variable, termed complete cross-validation (Kohavi, 1995). However, directly calculating this expectation requires averaging over all permissible partitions of S and is generally impractical since the number of such partitions grows exponentially with jSj. Therefore, it is customary to create an estimate by averaging the accuracies from a manageable number of partitions (Mitchell, 1997). Unfortunately, the variance of the random variable can be significant and obtaining an estimate with sufficiently low variance can require a large number of time-consuming trials. We present a technique for calculating the complete cross-validation (CCV) for the neare