V. Seferidis and M. Ghanbari, "General approachtoblock matching motion estimation," Optical Engineering, No. 7, pp. 1464--1474, Jul. 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....incorporated motion discontinuityinto the block based motion representation to obtain substantial quality improvement along moving object boundaries. The discontinuitybetween the motion vectors in adjacent blocks however still exists. Bergeron [1] Fuh [4] Papadopoulos [18] and Seferidis [20] [21] used more complex functions such as affine, perspective, bilinear, 2nd order polynomial transformations to represent the motion field inside each block. The schemes require more bits in the coding of motion field. As an example, an affine representation requires 3 vectors, instead of the ....

V. Seferidis and M. Ghanbari, "General approachtoblock matching motion estimation," Optical Engineering, No. 7, pp. 1464--1474, Jul. 1993.

....each block using block matching [2] To improve the performance of those coders, recent works [3, 4, 5] applied quadtree motion segmentation. A given block is split when its prediction fails to match some quality criterion. This way, the amount of side information is reduced. Other reported works [6, 7, 8] use higher order motion models to improve the prediction but keep using a fixed motion segmentation. In this work we propose a low bit rate coder which improves the performance of the standard H.261 coder by combining adaptive motion segmentation, improved motion estimation, and efficient ....

....we use 3 corresponding motion vectors. The regions in the current frame are selected to have shape of a rectangular block and the corner vectors are the motion vectors of 3 corners of the region. The corner vectors were used in other works for the actual estimation of the motion parameters (e.g. [7]) We use them here only for the purpose of representing the affine motion parameters in order to reduce their sensitivity to quantization errors, as explained below. The following relations exist between the corners of a block B and the corners of the parallelogram B 0 resulting from ....

V. Seferidis and M. Ghanbari. General approach to block-matching motion estimation. Optical Engineering, 32(7):1464--1474, July 1993.

....of the 3 D motion in the scene. So unless an object in the scene moves parallel to the image plane, its motion in the image will not be a 2 D translation. The 2 D motion estimation used in conventional video coding can therefore be improved upon by using more appropriate motion parameters. In [17] three types of 2 D transformations are used that model the projection of the 3 D motion: affine, perspective and bilinear transformation. It is shown that these motion estimations result in better compression ratios than just translation motion estimation. The parameters of the transformation are ....

V. Seferidis and M. Ghanbari. General approach to block-matching motion estimation. Optical Engineering, 32(7):1464--1474, July 1993.

....Also, illumination changes are mainly non uniform and the uncovered background in image coding applications cannot be easily ignored. To address the incompatibility between these unrealistic assumptions and practical motions, methods that employ spatial transformations have been proposed [4, 5, 6]. The Block Matching with a Geometric Transformation (BMGT) is capable of complex motion compensation with commendable bitrate improvement. This method is so efficient that most of the time only the motion vectors are required to be sent without the need for the error difference signal. In this ....

....of the BMGT technique, the image sequence is coded with only motion vectors. To improve compression further, motion compensation is applied on variable block sizes using image segmentation. In contrast to the continuous mesh concept used in [4, 6] in the presented method, as well as in [5], the optimal mappings are not constrained by the neighbouring mapped blocks. This feature makes it possible that a quadrilateral of any size and shape from the previous frame to be mapped into the current block, hence compensating for a more general complex motion. This part will be discussed in ....

[Article contains additional citation context not shown here]

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation, " Journal of Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

....parameters to each of these blocks. As the region described by one set of parameters is small, its motion can be efficiently represented by a simple model. For this reason, these techniques commonly consider a translational model (Eq. 6. 12) although a generalized block matching is proposed in [42] which supports more complex models. In [43] a block based differential motion estimation is proposed which relies on a hierarchy of models, namely models of various complexities which can be applied to each block. The difficulty of the approach is in measuring the performance of the respective ....

....Direct parametric motion estimation by means of a matching technique is proposed in [27, 46] This algorithm is a generalization of the classical block matching technique [7, 8, 9, 10] widely used in video coding. The motion parameters are computed by minimizing a disparity measure. Similar to [42], the motion model is no longer restricted to a translation as in the classical technique. Therefore, the generalized matching technique can handle more complex models such as the affine model (Eq. 6.11) Furthermore, the proposed generalized matching technique allows for regions of any shape, ....

[Article contains additional citation context not shown here]

V. Seferidis and M. Ghanbari. General approach to block-matching motion estimation. Optical Engineering, vol. 32, no. 7, pp. 1464-1474, July 1993.

....compression efficiency as well as object based interactivity and universal accessibility. MPEG 4 is expected to issue a draft international standard (DIS) by 1998. A variety of methods have been proposed in the literature for improved motion compensation (MC) including deformable block MC [2], overlapped block MC (OBMC) 3] and 2 D mesh based MC [4, 5, 6, 7, 8, 9, 10] Unlike the translational block model, mesh based motion compensation features a continuously varying motion field which can model local affine, perspective, or bilinear spatial deformations. Brusewitz [4] proposed ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation," Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

.... motion vectors at the node points such that they constitute a compact representation of the entire 2 D dense motion field (while still preserving the connectivity of the mesh) Estimation of the motion vectors around each node independently (e.g. by block matching or deformable block matching [8]) is usually not desirable because: i) motion vectors do not represent the entire 2 D dense motion field, and ii) motion vectors may cross each other (especially around small patches) destroying the connectivity of the mesh. Recently, search based solutions to node point motion estimation with ....

V. Seferidis and M. Ghanbari, "General approach to block matching motion estimation," Optical Engineering, pp. 1464-1474, 1993.

.... motion vectors at the node points such that they constitute a compact representation of the entire 2 D dense motion field (while still preserving the connectivity of the mesh) Estimation of the motion vectors around each node independently (e.g. by block matching or deformable block matching [8]) is usually not desirable because: i) motion vectors do not represent the entire 2 D dense motion field, and ii) motion vectors may cross each other (especially around small patches) destroying the connectivity of the mesh. Recently, search based solutions to node point motion estimation with ....

V. Seferidis and M. Ghanbari, "General approach to block matching motion estimation," Optical Engineering, pp. 1464-1474, 1993.

....this formulation is quite powerful, complexity of the resulting algorithm and possible convergence to a local minimum may hamper the effectiveness of the method. 3 2 D Object Modeling A variety of methods have been proposed for improved motion compensation (MC) including deformable block MC [15], overlapped block MC [16] region based MC [17, 18, 19] and 2 D mesh based MC [20, 21, 22, 23, 24, 25, 26, 27] These approaches can also be applied to 2 D objectbased MC given the boundaries of objects of interest. Among the region based approaches Hoetter [17] proposed the source model of ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation," Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

....objects, respectively. 19] and comparing the resulting displaced object difference (DOD) against a threshold. The displacement vector at the vertex points in two consecutive frames can be computed using either gradient based methods [20] block matching [21] or generalized block matching [22]. Generalized block matching methods that allow affine shape deformations and affine amplitude transformations in intensity [22] 23] have been used in this paper. More specifically, it is assumed that, within a neighboring region R, each object point p(x; y) at time t = t 1 maps to p 0 (x; y) ....

.... vector at the vertex points in two consecutive frames can be computed using either gradient based methods [20] block matching [21] or generalized block matching [22] Generalized block matching methods that allow affine shape deformations and affine amplitude transformations in intensity [22], 23] have been used in this paper. More specifically, it is assumed that, within a neighboring region R, each object point p(x; y) at time t = t 1 maps to p 0 (x; y) Mp(x; y) d at time t = t 2 by a spatial transformation, given by Mp(x; y) d = 0 s x Cos x Gammas y Sin y s x Sin ....

V. Seferidis and M. Ghanbari. General approach to block-matching motion estimation. Optical Engineering, 32:1464--474, 1993.

....grant to the Center for Electronic Imaging Systems at the University of Rochester. is because translational block model i) cannot handle image rotations and scaling, and ii) cannot preserve the neighboring relations between the blocks, resulting in blocking artifacts. Generalized block matching [6] was proposed to allow for spatial transformations (e.g. affine, perspective, etc. to overcome the limitations of the translational model. However, it does not address the discontinuity problem at the block boundaries. Overlapped block motion compensation, which is included in the advanced ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation," Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

....In fact, under the rigid body assumption, ie that moving bodies do not alter their shape, affine transformations are capable of perfectly modelling the motion resulting from orthographic 3 D to 2 D projection. The idea of using affine transformations for motion estimation purposes is not new [29, 30, 31]. Although it might be possible to determine the full affine transformation directly by search, this has proved computationally expensive [30] Most approaches in use attempts to solve the motion equations directly by adopting some feature based scheme [32, 33] which can involve the solution of as ....

....the motion resulting from orthographic 3 D to 2 D projection. The idea of using affine transformations for motion estimation purposes is not new [29, 30, 31] Although it might be possible to determine the full affine transformation directly by search, this has proved computationally expensive [30]. Most approaches in use attempts to solve the motion equations directly by adopting some feature based scheme [32, 33] which can involve the solution of as many as 15 simultaneous non linear equations. De Castro [34] used Fourier methods to estimate rotation. Adelson et al. 29] use a ....

V. Seferidis and M. Ghanbari, "General Approach to Block-Matching Motion Estimation," Optical Engineering, vol. 32, no. 7, pp. 1464--1474, 1993.

....motion model parameters within a region based framework. The general idea is to create an adaptive tessellation of the image plane and to perturb this tessellation to minimise an error measure over the regions. A generalised block matching algorithm was proposed by Seferidis and Ghanbari [123]. They estimate higher order model parameters relating local regions by deformation of the blocks. By moving the corners of the block according to the perspective transformation and transforming the underlying region using bilinear interpolation they then minimise the sum of squared differences ....

....frame and the untransformed region from the next. The full search approach in a higher 2.3 The Use of Higher order Models 43 order parameter space is computationally impractical. For the perspective transform, for a search space of SigmaN pixels (2N Gamma 1) 8 positions must be evaluated [123]. For this reason, they propose a range of optimisations, such as working within a multiresolution structure in a coarse to fine fashion and propose faster search strategies by what amounts to subsampling the perspective parameter space, but since it is still a matter of minimising a function over ....

[Article contains additional citation context not shown here]

V. Seferidis and M. Ghanbari. General approach to block-matching motion estimation. Optical Engineering, 32(7):1464--1474, 1993.

....structure to determine the global motion field. As such, it avoids having to rely on noise sensitive gradient estimates used in approaches based on optical flow fields [2] 3] 6] while at the same time avoiding the high computational costs associated with generalised block matching methods [4][5] The scheme is implemented within the framework of the multiresolution Fourier transform (MFT) a generalised form of wavelet transform which provides local spectral estimates over multiple scales [7] After outlining the underlying motion model, the estimation algorithm is described in ....

V.Seferidis and M.Ghanbari, "General approach to block-matching motion estimation", Optical Eng., 32, pp 1464-1474, 1993.

....be very costly, and despite this, H.263 uses it and still is more efficient than H.261. 2 Motion Compensation using Geometric Transformations Motion compensation efficiency can be further improved by using geometric transforms to deform moving quadrilateral patches to achieve subpixel accuracy [3]. This technique is so efficient that the motion compensated predicted (MCP) error for most parts of the picture are not needed to be coded at all [4] The main reason for using bilinear transform motion compensation than the conventional block matching for low bit rates can be given as follows: In ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation, " Journal of Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

....of transform is used in our BMST method. Similar techniques have been used for mesh based video coding systems [4,5] but the presented method is block based, where a quadrilateral in the previous frame is deformed to best match the current block, independent of the motion of the neighbour blocks [6]. Using Bilinear transformation, pixels in the UV plane are mapped into pixels in the XY plane by: X(u; v) a 0 a 1 u a 2 v a 3 uv (1) Y (u; v) b 0 b 1 u b 2 v b 3 uv (2) In a block based Bilinear transform, Fig. 1, a block in the current frame, ABCD, is matched against a set ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation, " Journal of Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

....Despite this, H.263 uses more overhead than H.261 and is still more efficient, because the coding gain achieved by better motion compensation offsets the extra overhead bits. Motion compensation can be further improved using geometric transformations to deform the moving quadrangular patches [6,7], achieving subpixel accuracy. This technique works so well that the Motion Compensated Predicted (MCP) errors in most parts of the pictures need not be coded. Having emphasised the importance of motion estimation in video compression efficiency, in this paper we introduce a very low bit rate ....

....one that minimises the mean squared or absolute distortions between the two blocks. Search for the best match starts by taking a block in the previous frame of the same size and shape to the current block (usually square) and then moving the coordinates of the quadrilateral one pixel at a time [6]. This deforms the quadrilateral to an arbitrary shape and after every new position of each vertex of the quadrilateral a new matching test has to be carried out. In this work we have used bilinear transform, which is a non uniform transform, hence can map a pixel into a group of pixels or ....

V. Seferidis and M. Ghanbari, "General approach to block-matching motion estimation," Journal of Optical Engineering, vol. 32, pp. 1464--1474, July 1993.

No context found.

V. Seferidsi and M. Ghanbari, "General Approach to Block-Matching Motion Estimation," Optical Engineering, 32(7), 1464--1474, (Jul. 1993).

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