In this paper, we propose a method for automatic, unequal and omni-directional anisotropic diffusion filtering of video sequences. Our method, which consists of automatic foreground/background discrimination using the color characteristics of the human face and skin, unequal filtering of the determined regions and then composition of the filtered regions into the final frames, is applied prior to encoding the video sequences. In addition to yielding higher visual quality of the decoded sequences than that of the decoded sequences with equal filtering over the entire frame, our method is robust with respect to skin tone and lighting variations, has low complexity and requires no calibration. We illustrate the main advantages of our method using videoconferencing sequences.

In this paper, we exploit the stylistic characteristics of high-quality entertainment movie sequences in terms of their textured background for coding purposes. More specifically, we propose a content-based coding method by texture replacement. At the encoder, texture is removed from selected regions of the original frames. The resulting frames with the texture removed and the parameters of the removed texture are then encoded. At the decoder, the boundaries of the regions without texture are identified and new texture, which is synthesized using the decoded texture parameters, is mapped onto these regions. Our experimental results confirm the main advantages of the proposed texture replacement method: significant bit rate reduction of the compressed movie sequences with the texture removed, and higher visual quality of the textured background regions in the decoded movie sequences with synthesized texture than that of the regions in the sequences simply encoded and decoded. Even more, our method can be applied as an overlay onto any standards-compliant coding system.

In this paper, we propose a method for direct mode enhancement in B pictures in the framework of H.264 (MPEG-4/Part 10), which employs reduction of outliers, clustering of the Lagrangian values and specification of smaller values of the Lagrangian multiplier in the rate-distortion optimization for mode selection. Experimental results using high quality video sequences show that bit rate reduction and no subjective visual loss in the decoded pictures are obtained using our method. Moreover, we verify that our method achieves further bit rate reduction without introducing visible distortion in the decoded sequences when other solutions would introduce inadmissible artifacts. Last but not least, the proposed method is a simple and useful add-on, which is applicable in any video coding system that employs rate-distortion optimization for coding mode selection.

We propose a method for texture replacement in video sequences. Our method, which is applied at the encoder side, consists of removal of texture from selected regions of the original frames, synthesis of new texture, and mapping of the new texture back onto the segmented regions. The texture removal stage employs highly effective color-based angular maps. The texture analysis and texture synthesis stages make use of steerable pyramids. The latter stage also employs constraints that are derived using a vocabulary and grammar for color pattern similarity evaluation that have been introduced previously. Because they have different characteristics than those of the original textures, the synthesized textures can be coded more effectively. Consequently and most importantly, significantly reduced bit rates of the compressed video sequences with the texture replaced are obtained as compared to those of the original sequences. Moreover, because the synthesized textures have similar perceptual characteristics to those of the original textures, the video sequences with the texture replaced are also visually similar to the original sequences. Even more, because it is performed at the encoder and it does not have any impact on the decoder, our texture replacement method is cost effective. We illustrate its performance and computational efficiency using movie sequences.

In this paper, we propose a method for texture replacement in video sequences at the encoder. Our method consists of removal of texture from selected regions of the original frames, analysis of the removed texture, synthesis of new texture, and mapping of the new texture back onto the segmented regions. Our proposed method yields significantly reduced bit rates of the compressed video sequences with the texture replaced as compared to those of the original sequences. Moreover, the video sequences with texture replacement are visually similar to the original sequences. Even more, our decoder-independent method is cost effective.

We propose a content-based method for coding entertainment movie sequences using texture replacement at the encoder and texture synthesis and mapping at the decoder. Our method reduces the bit rate of the compressed movie sequences significantly and yields higher visual quality of the textured background regions in the decoded movie sequences with mapped texture than that of the regions in the sequences simply encoded and decoded. Moreover, our method is efficient in terms of speed and may be applied as an overlay onto any standards-compliant coding system.

We propose a method for shape description of objects in color images. Our method employs angular maps to identify significant changes of color within the image, which are then used to drive snake models. Experimental results show that our angular map-driven snake method not only yields an accurate description of an object shape, but also it is computationally efficient.

We discuss parametric snake-based methods for the localization of object boundaries in color images. We illustrate the discussion using examples from our on-going work.

We propose a method for high-order image subsampling using feedforward artificial neural networks (FANNs). In our method, the high-order subsampling process is decomposed into a sequence of first-order subsampling stages. The first stage employs a tridiagonally symmetrical FANN, which is obtained by applying the design algorithm introduced by Dumitras and Kossentini (see IEEE Trans. Signal Processing, vol.48, p.1446-55, 2000). The second stage employs a small fully connected FANN. The algorithm used to train both FANNs employs information about local edges (extracted using pattern matching) to perform effective subsampling of both high detail and smooth image areas. We show that our multistage first-order subsampling method achieves excellent speed-performance tradeoffs, and it consistently outperforms traditional lowpass filtering and subsampling methods both subjectively and objectively.

We propose a method for shape description of objects in color images. Our method employs angular maps in order to identify significant changes of color within the image, which are then used to drive snake models. To obtain an angular map, the angle values of the vectors corresponding to color image pixels are first computed with respect to a reference vector, and organized in a two-dimensional matrix. To identify significant color changes within the original image, the edges of the angular map are next extracted. The resulting edge map is then presented to a snake model. Distance and gradient vector flow snake models have been employed in this work. Experimental results show, not only that the resulting object shape descriptions are accurate and quite similar, but also that our method is computationally efficient and flexible.

We propose a method for high-order image subsampling using feedforward artificial neural networks (FANNs). Our method employs a tridiagonally symmetrical FANN, which is obtained by applying the design algorithm proposed by Dumitraa and Kossentini (see IEEE Transactions on Signal Processing, vol.48, no.5, p.1446-55, 2000), and a small, fully connected FANN. We show that our subsampling method provides excellent speed-performance tradeoffs as compared to those of the method based exclusively on fully connected FANNs.

We propose a method for image subsampling using feedforward artificial neural networks (FANN). Our method employs pattern matching in order to extract information about the presence of local edges. This information is then used to select the FANN desired output values during training. We show that, when applied to first-order and high-order image subsampling, our method effectively subsamples both high detail and smooth image areas, consistently outperforming traditional lowpass filtering and subsampling (LPFS) methods.

We introduce a fast and high performance image subsampling method using feedforward artificial neural networks (FANNs). Our method employs a pattern matching technique to extract local edge information within the image, in order to select the FANN desired output values during the supervised training stage. Subjective and objective evaluations of experimental results using still images and video frames show that our method, while less computationally intensive, outperforms the standard lowpass filtering and subsampling method.

In this paper, we present a video chrominance subsampling method using feedforward artificial neural networks (FANNs). Experimental results show that our method outperforms spatial subsampling obtained via low pass filtering and decimation both objectively and subjectively. Other advantages of our algorithm are computational efficiency and low memory requirements. Moreover, no pre- or post-processing is required by our method.

We prove that a multi-objective optimization approach of the learning process in feedforward neural networks leads to better signal processing performances in the presence of impulsive noise and outliers.