Amir Averbuch

Image compression using wavelet transform and multiresolution decomposition

Schemes for image compression of black-and-white images based on the wavelet transform are presented. The multiresolution nature of the discrete wavelet transform is proven as a powerful tool to represent images decomposed along the vertical and horizontal directions using the pyramidal multiresolution scheme. The wavelet transform decomposes the image into a set of subimages called shapes with different resolutions corresponding to different frequency bands. Hence, different allocations are tested, assuming that details at high resolution and diagonal directions are less visible to the human eye. The resultant coefficients are vector quantized (VQ) using the LGB algorithm. By using an error correction method that approximates the reconstructed coefficients quantization error, we minimize distortion for a given compression rate at low computational cost. Several compression techniques are tested. In the first experiment, several 512x512 images are trained together and common table codes created. Using these tables, the training sequence black-and-white images achieve a compression ratio of 60-65 and a PSNR of 30-33. To investigate the compression on images not part of the training set, many 480x480 images of uncalibrated faces are trained together and yield global tables code. Images of faces outside the training set are compressed and reconstructed using the resulting tables. The compression ratio is 40; PSNRs are 30-36. Images from the training set have similar compression values and quality. Finally, another compression method based on the end vector bit allocation is examined.

Automatic segmentation of moving objects in video sequences: a region labeling approach

The emerging video coding standard MPEG-4 enables various content-based functionalities for multimedia applications. To support such functionalities, as well as to improve coding efficiency, MPEG-4 relies on a decomposition of each frame of an image sequence into video object planes (VOP). Each VOP corresponds to a single moving object in the scene. This paper presents a new method for automatic segmentation of moving objects in image sequences for VOP extraction. We formulate the problem as graph labeling over a region adjacency graph (RAG), based on motion information. The label field is modeled as a Markov random field (MRF). An initial spatial partition of each frame is obtained by a fast, floating-point based implementation of the watershed algorithm. The motion of each region is estimated by hierarchical region matching. To avoid inaccuracies in occlusion areas, a novel motion validation scheme is presented. A dynamic memory, based on object tracking, is incorporated into the segmentation process to maintain temporal coherence of the segmentation. Finally, a labeling is obtained by maximization of the a posteriori probability of the MRF using motion information, spatial information and the memory. The optimization is carried out by highest confidence first (HCF). Experimental results for several video sequences demonstrate the effectiveness of the proposed approach.

Fast adaptive wavelet packet image compression

Wavelets are ill-suited to represent oscillatory patterns: rapid variations of intensity can only be described by the small scale wavelet coefficients, which are often quantized to zero, even at high bit rates. Our goal is to provide a fast numerical implementation of the best wavelet packet algorithm in order to demonstrate that an advantage can be gained by constructing a basis adapted to a target image. Emphasis is placed on developing algorithms that are computationally efficient. We developed a new fast two-dimensional (2-D) convolution decimation algorithm with factorized nonseparable 2-D filters. The algorithm is four times faster than a standard convolution-decimation. An extensive evaluation of the algorithm was performed on a large class of textured images. Because of its ability to reproduce textures so well, the wavelet packet coder significantly out performs one of the best wavelet coder on images such as Barbara and fingerprints, both visually and in term of PSNR.

Color image segmentation based on adaptive local thresholds

Abstract The goal of still color image segmentation is to divide the image into homogeneous regions. Object extraction, object recognition and object-based compression are typical applications that use still segmentation as a low-level image processing. In this paper, we present a new method for color image segmentation. The proposed algorithm divides the image into homogeneous regions by local thresholds. The number of thresholds and their values are adaptively derived by an automatic process, where local information is taken into consideration. First, the watershed algorithm is applied. Its results are used as an initialization for the next step, which is iterative merging process. During the iterative process, regions are merged and local thresholds are derived. The thresholds are determined one-by-one at different times during the merging process. Every threshold is calculated by local information on any region and its surroundings. Any statistical information on the input images is not given. The algorithm is found to be reliable and robust to different kind of images.

Pseudopolar-based estimation of large translations, rotations, and scalings in images

One of the major challenges related to image registration is the estimation of large motions without prior knowledge. This work presents a Fourier-based approach that estimates large translations, scalings, and rotations. The algorithm uses the pseudopolar (PP) Fourier transform to achieve substantial improved approximations of the polar and log-polar Fourier transforms of an image. Thus, rotations and scalings are reduced to translations which are estimated using phase correlation. By utilizing the PP grid, we increase the performance (accuracy, speed, and robustness) of the registration algorithms. Scales up to 4 and arbitrary rotation angles can be robustly recovered, compared to a maximum scaling of 2 recovered by state-of-the-art algorithms. The algorithm only utilizes one-dimensional fast Fourier transform computations whose overall complexity is significantly lower than prior works. Experimental results demonstrate the applicability of the proposed algorithms.

Experiments with the Tangora 20,000 word speech recognizer

The Speech Recognition Group at IBM Research in Yorktown Heights has developed a real-time, isolated-utterance speech recognizer for natural language based on the IBM Personal Computer AT and IBM Signal Processors. The system has recently been enhanced by expanding the vocabulary from 5,000 words to 20,000 words and by the addition of a speech workstation to support usability studies on document creation by voice. The system supports spelling and interactive personalization to augment the vocabularies. This paper describes the implementation, user interface, and comparative performance of the recognizer.

Fast gradient methods based on global motion estimation for video compression

This paper presents a fast global motion estimation (GME) algorithm based on gradient methods (GM), which can be used for real-time applications, such as in MPEG4 video compression. This approach improves the existing state-of-the-art GME algorithms by introducing two major modifications: first, only a small subset (down to 3%) of the original image pixels is used in the estimation process. Second, an interpolation-free formulation of the basic GM is derived, further decreasing the computational complexity. Experimental results show no loss of GME accuracy and compression efficiency compared to the MPEG-4 verification model, while reducing the computational complexity of the GME by a factor of 20.

Digital image thresholding, based on topological stable-state

A new approach for image segmentation for scenes that contain distinct objects is presented. A sequence of graphs Ns(t) is defined, where Ns(t) is the number of connected objects composed of at least s pixels, for the image thresholded at t. The sequence of graphs is built in almost linear time complexity, namely at O(@a(n, n). n), where @a(n, n) is the inverse of the Ackermann function, and n is the number of pixels in the image. Stable states on the graph in the appropriate ""resolution"" s^* correspond to threshold values that yield a segmentation similar to a human observer. The relevance of a Percolation model to the graphs Ns(t) is discussed.

Radar target tracking-Viterbi versus IMM

The performance of the Viterbi and the interacting multiple model (IMM) algorithms, applied to radar tracking and detection, are investigated and compared. Two different cases are considered. In the first case, target acceleration is identical to one of the system models while in the second case it is not. The performance of the algorithms depends monotonically on the maximal magnitude of the difference between models acceleration, on the time interval between measurements, and on the reciprocal of the measurements error standard deviation. In general, when these parameters are relatively large, both algorithms perform well. When they are relatively small, the Viterbi algorithm is better. However, during delay periods, namely right after the start of a maneuver, the IMM algorithm provides better estimations. >

Multilayered image representation: application to image compression

The main contribution of this work is a new paradigm for image representation and image compression. We describe a new multilayered representation technique for images. An image is parsed into a superposition of coherent layers: piecewise smooth regions layer, textures layer, etc. The multilayered decomposition algorithm consists in a cascade of compressions applied successively to the image itself and to the residuals that resulted from the previous compressions. During each iteration of the algorithm, we code the residual part in a lossy way: we only retain the most significant structures of the residual part, which results in a sparse representation. Each layer is encoded independently with a different transform, or basis, at a different bitrate, and the combination of the compressed layers can always be reconstructed in a meaningful way. The strength of the multilayer approach comes from the fact that different sets of basis functions complement each others: some of the basis functions will give reasonable account of the large trend of the data, while others will catch the local transients, or the oscillatory patterns. This multilayered representation has a lot of beautiful applications in image understanding, and image and video coding. We have implemented the algorithm and we have studied its capabilities.