V. Dang, A. Mansouri, and J. Konrad, "Motion estimation for region-based video coding", IEEE ICIP, pp. 189--192, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....So, it requires a regularization by imposing smoothness or uniformity constraints on the local motion field. In these methods, the Markov Random field (MRF) is often used to model images, and the smoothness constraint is expressed as an energy function of the associate Gibbs distribution [11] [12]. They set up an objective function as a sum of the energy of motion estimation error and the weight smoothness constraint. Based on the maximum a posterior (MAP) criterion, motion is then estimated by searching the motion field minimizing the objective function [13] 14] Drawbacks of this model ....

V. Dang, A. Mansouri, and J. Konrad, "Motion estimation for region-based video coding", IEEE ICIP, pp. 189--192, 1995.

....real video scene. A relatively smooth motion field is obtained within each object. The contour information can also be encoded efficiently by means of temporal updating using the linked object labels. This alleviates the bottleneck created by the contour information in region based coding schemes [7] where contours are coded frame by frame. The rest of this paper is organized as follows. The joint motion estimation segmentation algorithm based on the MRF model is presented in detail in section 2. We compress the motion fields by fitting an affine parametric model to each object. The ....

V Dang, A Mansouri, and J Konrad, "Motion estimation for region-based video coding", in Proc. IEEE Int. Conf. Image Processing. 1995, pp. 189-- 192, Washington, DC.

....defined in terms of intensity or texture as pointed out earlier. However, in our novel approach, the object motion and shape are guided by the spatial color intensity information, thus utilizing the observation that in an image sequence motion boundaries are a subset of the intensity boundaries [22, 23]. This not only improves the motion estimation segmentation process itself in extracting meaningful objects true to the scene, but also aids the process of coding the object intensities (especially in Intra mode) because a given object has a certain spatial cohesiveness as well. The MRF ....

V. Dang, A. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, pp. 189--192, Washington, DC, 1995.

....the motion failure regions. In contrast to the above top down approaches, there also exist bottomup procedures which start with many small subregions and group them together according to some merging criterion to form segments. A good example is the region based methods of Konrad et al. [34, 35], where the basic assumption is that the motion boundaries coincide with spatial intensity boundaries. In their scheme, a set of affine mapping parameters are found for each intensity region at time t, and neighboring regions with similar motion are fused. The affine parameters are calculated ....

....and the majority filter [44] are good examples. An alternative lossy compression method includes polygonal and or spline representation of the object boundaries [45, 46] For image sequences, one can always perform the above mentioned 2 D contour coding techniques on each frame independently [35]. Recently, methods to apply motion compensation to the contour coding have been proposed to further reduce the bit rate [44, 39, 47] However, as mentioned earlier, these algorithms find the temporally linked objects using only the intensity information, without regard to the motion. The motion ....

[Article contains additional citation context not shown here]

V. Dang, A. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, pp. 189--192, Washington, DC, 1995.

.... videoconferencing images at QCIF (176 Theta144) resolution: a) Carphone (frame 171) and (b) Miss America (frame 6) In general, image partitioning based on fixed rectangular blocks under translations is outperformed, in terms of intensity color residual, by region based affine motion models [20]. While the former is simple to implement in hardware, the latter requires fairly sophisticated image analysis. As a compromise, partitioning methods based on rectangular variable size blocks have been investigated [14] 70] 26] In such an approach, block size is locally reduced wherever a ....

....Fig. 5. Typical motion fields computed from sequence Carphone (Fig. 4. a) for different regions of support: a) blockbased (16 Theta16 blocks) b) pixel based (globally smooth as in (17) and (c,d) region based with affine motion model (Table I) For details of the region based algorithm, see [20]. closely delineates object boundaries. The impact of various regions of support on motion compensated prediction and prediction error is shown in Fig. 6. Note the blocking artifacts for the block based motion model and the associated 4dB penalty in the peak prediction error as compared with the ....

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, vol. II, pp. 189--192, Oct. 1995.

No context found.

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, vol. II, pp. 189--192, Oct. 1995.

....stress that both the boundary complexity and the number of regions have to be minimized. The reduction of N may be done through the pairwise merging of regions, i.e. comparison of the total cost before and after the merge. We have carried out some preliminary investigations using the above models [10]. Figs. 4.3(e) and 4.3(f) show the motion and segmentation estimates for affine model (4.28) using both minimizations above except that inter region smoothing was disabled (V sim =0) Note the improved motion estimate in the car window compared to pixel and block based estimates. Also, the ....

....model (4.28) using both minimizations above except that inter region smoothing was disabled (V sim =0) Note the improved motion estimate in the car window compared to pixel and block based estimates. Also, the prediction error energy is significantly lower for the region based motion estimate [10]. From the coding point of view, however, the gain is not obvious since the rate needed for motion attributes (p and s) in the case of region based approach is higher than that needed to transmit block displacements. Clearly, to minimize the overall rate, a judicious allocation of rate between ....

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for regionbased video coding," in Proc. IEEE Int. Conf. Image Processing, pp. II.189-- II.192, Oct. 1995.

....For a large circular region they become Bessel functions separable in polar coordinates. 4. SIMULATION RESULTS We have applied the proposed transform to a frame from the sequence carphone . Figure 2 shows the original image. We used a minimum description length segmentation as proposed in [3]. This segmentation is based on intensity and motion, and results in 10 regions as depicted in Figure 3. A region based approximation of the image with restricted basis is shown in Figure 4. It yields a PSNR of 34.2 dB using 15 of the basis vectors. Figure 2: Frame # 3 of carphone Figure 3: ....

....in 10 regions as depicted in Figure 3. A region based approximation of the image with restricted basis is shown in Figure 4. It yields a PSNR of 34.2 dB using 15 of the basis vectors. Figure 2: Frame # 3 of carphone Figure 3: Intensity and motion based segmentation consisting of 10 regions [3] Figure 4: Approximation using 15 of coefficients (34.2 dB) 5. CONCLUSIONS We have proposed a new approach to construct a linear transform for arbitrarily shaped image segments. Instead of selecting cutouts from standard block transforms, the proposed transform is based upon a stochastic model ....

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," Proc. IEEE Int. Conf. Image Process., Oct. 1995.

.... videoconferencing images at QCIF (176 Theta144) resolution: a) Carphone (frame 171) and (b) Miss America (frame 6) In general, image partitioning based on fixed rectangular blocks under translations is outperformed, in terms of intensity color residual, by region based affine motion models [20]. While the former is simple to implement in hardware, the latter requires fairly sophisticated image analysis. As a compromise, partitioning methods based on rectangular variable size blocks have been investigated [14, 69, 26] In such an approach, block size is locally reduced wherever a smaller ....

....Figure 5: Typical motion fields computed from sequence Carphone (Fig. 4. a) for different regions of support: a) block based (16 Theta16 blocks) b) pixel based (globallysmooth as in (17) and (c,d) region based with affine motion model (Table 1) For details of the region based algorithm, see [20]. as perceived by humans, it nevertheless closely delineates object boundaries. The impact of various regions of support on motion compensated prediction and prediction error is shown in Fig. 6. Note the blocking artifacts for the block based motion model and the associated 4dB penalty in the peak ....

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, vol. II, pp. 189--192, Oct. 1995.

....segmentation procedure is twofold: it must provide a rateefficient partitioning of images and it should result in regions that are meaningful for human observers; this would allow browsing of bit streams without complete decoding. In this paper, we extend our earlier approach to video segmentation [1] that was based on the premise that motion discontinuities coincide with intensity discontinuities. This premise was exploited to assure that final spatio temporal segmentation be closely related to region boundaries of a spatial segmentation. As we have demonstrated in [1] this premise leads to ....

....to video segmentation [1] that was based on the premise that motion discontinuities coincide with intensity discontinuities. This premise was exploited to assure that final spatio temporal segmentation be closely related to region boundaries of a spatial segmentation. As we have demonstrated in [1], this premise leads to partitions that are meaningful for human observers. A different approach to video segmentation based on the same premise has been also proposed in [3] Our approach is based on two steps: region fusion and region boundary adjustment. We extend our earlier approach in three ....

[Article contains additional citation context not shown here]

V.-N. Dang, A.-R. Mansouri, and J. Konrad, "Motion estimation for region-based video coding," in Proc. IEEE Int. Conf. Image Processing, pp. II.189--II.192, Oct. 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC