Fast tree-structured nearest neighbor encoding for vector quantization

Low-Complexity Hand Gesture Recognition System for Continuous Streams of Digits and Letters

In this paper, we propose a complete gesture recognition framework based on maximum cosine similarity and fast nearest neighbor (NN) techniques, which offers high-recognition accuracy and great computational advantages for three fundamental problems of gesture recognition: 1) isolated recognition; 2) gesture verification; and 3) gesture spotting on continuous data streams. To support our arguments, we provide a thorough evaluation on three large publicly available databases, examining various scenarios, such as noisy environments, limited number of training examples, and time delay in system's response. Our experimental results suggest that this simple NN-based approach is quite accurate for trajectory classification of digits and letters and could become a promising approach for implementations on low-power embedded systems.

Association of different prediction methods for determination of the efficiency and selectivity on neuron-based sensors

A technique has been developed to determine the efficiency and the selectivity of a single neuron-based sensor in identifying the nature of the chemical agents in an unknown sample. This has been achieved by exploiting the unique electrical identifiers, also known as "signature patterns", generated by the neuronal cell membrane. These were generated based on the variations to the extracellular electrical activity, due to the effect of a broad range of chemical agents. We demonstrate the prediction capability of the sensor in identifying the nature of an unknown test sample from a combination of three chemical agents, namely, ethanol, pyrethroid, and hydrogen peroxide. This was achieved through a two-step process. The first step was experimentally achieved by in situ recording of the changes to the extracellular electrical activity from the sensing sites or the array of microelectrodes that form the platform for patterning neurons. Simultaneous optical characterization of the cell array during the sensing process was performed to identify the associated physiological changes. The second step was mathematical and was based on developing a library of signature patterns for a set of concentrations of the various combinations of the three chemical agents. Two variants of the nearest neighbor algorithm scheme - (a) partial distance search method, and (b) search tree method, were implemented for the accurate detection of all the components with varying concentrations in the test samples of unknown nature. This technique exhibits reliability in identification up to parts-per-billion (ppb) sensitivity. The capability of standardization of this technique for potential commercial applications is also discussed.

Color object indexing and retrieval in digital libraries

In our previous work, illumination invariant object recognition was achieved by normalizing the three color bands. We further employed the compressed histogram of the chromaticity to arrive at a valuable representation of an object which can facilitate high retrieval accuracy. The first shortcoming of this method lies in the usage of a uniform quantization scheme in obtaining the chromaticity, which is not in agreement with the perception of the human vision system. In this paper, we develop an approach using the CIE UCS transform to circumvent this problem. Second, instead of using uncompressed images to achieve the illumination invariant indexing and retrieval, we carry out our indexing process directly in the DCT domain by using several coefficients from each macro-block. Third, in light of the special properties of the normalized chromaticity histogram frames, the foundation of the ensuing low-pass filtering, an additional step is inserted to render this frame smoother thus resulting in a better data reduction. Fourth, in order to facilitate efficient retrieval during the data query phase, which is of utmost importance in digital libraries, the 36-dimensional model vectors as the indices of model images in digital libraries are clustered by use of vector quantization techniques. This clustering strategy reduces the searching space by an order of magnitude. Desirable results have been observed from our experiments using the proposed color-object-indexing/retrieval algorithm.

S-TREE: self-organizing trees for data clustering and online vector quantization

This paper introduces S-TREE (Self-Organizing Tree), a family of models that use unsupervised learning to construct hierarchical representations of data and online tree-structured vector quantizers. The S-TREE1 model, which features a new tree-building algorithm, can be implemented with various cost functions. An alternative implementation, S-TREE2, which uses a new double-path search procedure, is also developed. The performance of the S-TREE algorithms is illustrated with data clustering and vector quantization examples, including a Gauss-Markov source benchmark and an image compression application. S-TREE performance on these tasks is compared with the standard tree-structured vector quantizer (TSVQ) and the generalized Lloyd algorithm (GLA). The image reconstruction quality with S-TREE2 approaches that of GLA while taking less than 10% of computer time. S-TREE1 and S-TREE2 also compare favorably with the standard TSVQ in both the time needed to create the codebook and the quality of image reconstruction.

Fast fractal compression of greyscale images

A new algorithm for fractal compression of greyscale images is presented. It uses some previous results allowing the compression process to be reduced to a nearest neighbors problem, and is essentially based on a geometrical partition of the image block feature space. Experimental comparisons with previously published methods show a significant improvement in speed with no quality loss.