Andre Zaccarin

New fast algorithms for the estimation of block motion vectors

Two algorithms for block motion estimation that produce performance similar to that of exhaustive search but with computation reduced by a factor of 8 or 16 are presented. The algorithms are based on motion-field and pixel subsampling. A subsampled motion field is first determined by estimating the motion vectors for a fraction of the blocks. The motion vectors for these blocks are determined by using only a fraction of the pixels at any searched location and by alternating the pixel subsampling patterns with the searched locations. The subsampled motion field is then interpolated so that a motion vector is determined for each block of pixels. The algorithms are more robust than previously proposed fast algorithms and both can easily be combined with a hierarchical search strategy. One of the algorithms is fully compatible with MPEG-I. >

A novel approach for coding color quantized images

An approach to the lossy compression of color images with limited palette that does not require color quantization of the decoded image is presented. The algorithm is particularly suited for coding images using an image-dependent palette. The technique restricts the pixels of the decoded image to take values only in the original palette. Thus, the decoded image can be readily displayed without having to be quantized. For comparable quality and bit rates, the technique significantly reduces the decoder computational complexity.

Automated seed detection and three-dimensional reconstruction. II. Reconstruction of permanent prostate implants using simulated annealing

We present an algorithm, based on simulated annealing, for automatic seed matching and three-dimensional spatial coordinate reconstruction using either three radiographic films or three fluoroscopic images taken from different perspectives. The matching problem is defined in the framework of combinatorial optimization, which allows robust reconstruction in presence of calibration imprecision, patient movements, and isometric distortions. Furthermore, by using a global criterion to select the correct matching, we evade common problems of the three-film method and its variants in presence of noise. The algorithm has been tested on 112 clinical cases and 100 simulated implants and used clinically on more than 100 cases. Simulated implants were reconstructed with an average error of 0.21 mm. For clinical cases, comparison of the precision is performed between results obtained with this new method and results obtained using the three-film technique. Compared to the latter technique, the reconstruction precision was improved in 62% of the clinical cases.

Automated seed detection and three-dimensional reconstruction. I. Seed localization from fluoroscopic images or radiographs

An automated procedure for the detection of the position and the orientation of radioactive seeds on fluoroscopic images or scanned radiographs is presented. The extracted positions of seed centers and the orientations are used for three-dimensional reconstruction of permanent prostate implants. The extraction procedure requires several steps: correction of image intensifier distortions, normalization, background removal, automatic threshold selection, thresholding, and finally, moment analysis and classification of the connected components. The algorithm was tested on 75 fluoroscopic images. The results show that, on average, 92% of the seeds are detected automatically. The orientation is found with an error smaller than 50 for 75% of the seeds. The orientation of overlapping seeds (10%) should be considered as an estimate at best. The image processing procedure can also be used for seed or catheter detection in CT images, with minor modifications.

Fast algorithms for block motion estimation

Two fast algorithms are presented to reduce the computational complexity of block motion estimation. Previously proposed fast algorithms reduce the number of computations by limiting the number of locations that are searched to find the motion vectors of each block. The techniques presented use block and pixel subsampling to reduce the number of computations of block motion estimation. Because an exhaustive search is used to estimate a fraction of the motion vectors, these techniques are more robust than those previously proposed. Alternating the pixel subsampling patterns with the searched locations minimizes the effects of the aliasing that is introduced by subsampling the pixels without prior low-pass filtering. Block and pixel subsampling by a factor of four each reduce the computational complexity of block motion estimation by a factor of 16. Performance similar to full exhaustive search is obtained.<<ETX>>

A system for reliable dissolve detection in videos

Automatic shot boundary detection has been an active research area for nearly a decade and has led to high performance detection algorithms for hard cuts, fades and wipes. Reliable dissolve detection, however, is still an unsolved problem. We present the first robust and reliable dissolve detection system. A detection rate of 75% was achieved while reducing the false alarm rate to an acceptable level of 16% on a test video set for which so far the best reported detection and false alarm rate had been 66% and 59%, respectively. In addition, a dissolves temporal extent is estimated, too. The core ideas of our novel approach are firstly the creation of a dissolve synthesizer capable of creating in principle an infinite number of dissolve examples of any duration from a video database of raw video footage allowing us to use an advanced machine learning algorithm such as neural networks and support vector machines which require large training sets, secondly, two simple features capturing the characteristics of dissolves, thirdly, a fully temporal multi-resolution search based on a fixed position and fixed-scale transition/special effect detector enabling us to determine also the true duration of detected dissolves, and finally, a post processing step which uses global motion estimation to further reduce the number of falsely detected dissolves.

Automatic setup deviation measurements with electronic portal images for pelvic fields

The purpose of this work was to develop a fully automatic tool for the detection of setup deviation for small pelvic field using, in external beam radiotherapy, an electronic portal imaging device (EPID). The algorithm processes electronic portal images of prostate cancer patients. No fiducial points or user interventions are needed. Deviation measurements are based on bone edge detection performed with Laplacian of a Gaussian (LoG) operator. Two bone edge images are then correlated, one of which is a reference image taken as the first fraction image for the purpose of this study. The electronic portal images (EPI) also show band artefacts which are removed using the morphological top-hat transform. The algorithm was first validated with 59 phantom images acquired in clinical treatment conditions with known displacements. The algorithm was then validated with 79 clinical images where bone contours were delineated manually. For the phantom images, the setup deviations were measured with a absolute mean error of 0.59 mm and 0.47 mm with a standard deviation of 0.64 mm and 0.42 mm, horizontally and vertically, respectively. A second validation was performed using clinical prostate cancer images. The measured patient displacements have an absolute mean error of 0.48 mm and 1.41 mm with a standard deviation of 0.58 mm and 1.30 mm in the X and Y directions, respectively. The algorithm execution time on a SUN workstation is 5 s. This algorithm shows good potential as a setup deviation measurement tool in clinical practice. The possibility of using this algorithm combined with decision rules based on statistical observations is very promising.

Transform coding of color images with limited palette size

An approach for the compression of color images with limited palette size that does not require color quantization of the decoded image is presented. The technique restricts the pixels of the decoded image to take values only in the original palette. Thus, the decoded image can be readily displayed without having to be quantized. Results obtained with a typical image are included to compare a conventional coding scheme to the proposed one. For comparable quality and bit rates, the proposed technique significantly reduces the decoder computational complexity.<<ETX>>

Multiscale autoregressive image representation for texture segmentation

Texture segmentation or modeling plays an important role in image segmentation. In this paper, we investigate multiscale autoregressive representation for texture modeling and segmentation. The proposed algorithm also uses a multiresolution decomposition of an image, and fits an AR model to each image of the multiresolution pyramid. The set of AR coefficient vectors, one at each level, defines a model for a texture and this model is used as a predictor for the segmentation process. AR coefficient vectors are used to generate a prediction of the image pyramid, from which the prediction of the image to model is built. The resulting prediction error is used to discriminate textures in a segmentation algorithm. In the proposed structure, feedback can be included between pyramid levels by adding the prediction error at he previous level to the current level before an AR model fitting. M-AR can therefore be used as a predictor like an AR model. This is different from previous multiscale approaches for which data is used at each scale for the segmentation. Since we do not need to link data from different scales, this simplifies model processing for segmentation. The estimation error of the proposed multiscale AR approach has lower variance than that of an AR model, and is less correlated. Segmentation results also show M-AR to be an improvement to AR modeling.

Multi-rate encoding of a video sequence in the DCT domain

In todays streaming servers, video sequences are offered to users at different fixed bit rates. This paper presents an efficient approach for simultaneous encoding of a video sequence at multiple bit rates. In this encoder, motion estimation is performed only once for the reference stream. We also take the DCT out of the encoding loop so that it is only computed once per frame. As a result, no iDCT has to be computed at the encoder. However, motion compensation is performed in the DCT domain, and drift error can be introduced adaptively to reduce the computational cost of the DCT-domain motion compensation. Results show that significant computational reduction can be achieved with less than 0.3 dB loss in PSNR when compared to independently encoding the video sequence at multiple rates.