Search

There are many error concealment techniques for image processing. In the paper, the focus is on restoration of image with missing blocks or macroblocks. Different methods can be optimal for different kinds of images. In recent years, great attention was dedicated to textures, and specific methods were developed for their processing. Many of them use classification of textures as an integral part. It is also of an advantage to know the texture classification to select the best restoration technique. In the paper, selection based on texture classification with advanced local binary patterns and spatial distribution of dominant patterns is proposed. It is shown, that for classified textures, optimal error concealment method can be selected from predefined ones, resulting then in better restoration. For testing, three methods of extrapolation and texture synthesis were used.

In this paper a scheme for utilizing shape independent basis functions for a hierarchical multiresolution image compression is shown. First, an image is segmented and its segments' boundaries are polygon approximated, thus achieving an image mask. Second, this image mask and the image are used as an input of a three-level hierarchical encoder. The hierarchical encoder subsamples the image and the image mask and encodes them shape independently; it produces an output bit stream on a respective level that is also used on lower level(s) for further coding. On the base level a triangulation of the image mask is performed for superior performance. Another compression mode is, hence, introduced for the shape independent transform coding.

The submitted paper deals with separate scanning of individual image segments. A new image processing approach based on image segmentation and segment scanning is presented. The resulting individual segments 1-dimensional representation provides higher neighbor pixel similarity than the 1-dimensional representation of the original image. This increased adjacent pixel similarity was achieved even without application of different recursive 2-dimensional scanning methods [4], such as Peano-Hilbert scanning method [1]. The resulting 1-dimensional image representation provides a good base for applying lossless compression methods, such as the entropic coding. The paper contains also results analysis of the traditional method scanned segment pixels and adjacent pixel differences from the entropy point of view. As these results indicate the lossy compression methods could be applicable using this approach as well and might improve the final results as confirmed by simple prediction algorithm results presented in this paper. More complex and sophisticated lossy compression algorithms application will be a part of the future work.

In this paper, a novel adaptive error concealment (EC) algorithm, which lowers the requirements for channel coding, is proposed. It conceals errors in block-based image coding systems by using neural network. In this proposed algorithm, only the intra-frame information is used for reconstruction of the image with separated damaged blocks. The information of pixels surrounding a damaged block is used to recover the errors using the neural network models. Computer simulation results show that the visual quality and the MSE evaluation of a reconstructed image are significantly improved using the proposed EC algorithm. We propose also a simple non-neural approach for comparison.

Our approach to image region approximation offers a complete schemeconsisting of several steps. We separate the encoding of regionboundaries from the texture description. The original image is firstsegmented using an unsupervised segmentation method for color-textureregions. Following polygonal approximation of created regions causesthe degradation of region boundaries. The triangulation is then appliedto polygons and either all short edges, or all small triangles arefiltered out from the triangular mesh. This results in new smallerregions. The encoding and decoding of polygons and triangles is veryefficient, because we need to store only the vertices. For textureapproximation we use 2D shape independent orthogonal transforms (e.g.DCT II). The texture is encoded with a code similar to JPEG arithmeticcode. The encoding scheme proposed in this paper is much faster thanlatest approaches with polygonal approximation. We present the twotriangulation algorithms - constrained greedy (CGT) and constrainedDelaunay triangulation (CDT). Both CGT and CDT provide similar results.

In non-stationary channels, error rates vary considerably. The papercompares Yao's Adaptive Go-back-N (GBN) Automatic-Repeat-Request (ARQ)scheme with Adaptive go-back-N with sliding window mechanism which bothestimate the channel state in a simple manner, and adaptively switchtheir operation mode. The throughput of these schemes is compared inconditions of Additive White Gauss Noise (AWGN) channel withindependent errors using 16-QAM modulation.

In the contribution we develop a new method for object-orientedinterpolation of images. It is an important tool in image processing,since using interpolation we can considerably decrease amount of data,necessary for image reconstruction. Application of this interpolationenables to down-sample the object separately. The selected object canbe processed at the different sampling level. This approach allowsobject oriented zoom, for example. Moreover, object - oriented approachis a very novel idea that helps to understand the content of image.Method is created from cosine approximation implemented to coding withshape - independent basis functions.

In image processing and transmission, interpolation andextrapolation are of great importance whenever missing pixels have tobe filled in, and many methods have been proposed to solve thisproblem. In this paper we present a method for extrapolating themissing data with an existing set of basis functions of a selectedorthogonal transform. The best extrapolation is found according tolinear approximation theory as a weighted sum of basis functions, wherecoefficients of the sum are solutions of the derived matrix equation.

In this paper we analyze the adaptation rule, which estimates thechannel state and switches between hybrid ARQ(automatic-repeat-request) and pure ARQ. Convolutional code was chosenas FEC (forward-error-correction) in hybrid ARQ part and go-back-N ARQscheme is used in both cases. The adaptation rule is based on countingACKs and NAKs and its throughput analysis is made.

In non stationary channels, error rates vary considerably. This paper proposes an effective go-back-N (GBN) Automatic-Repeat-Request (ARQ) scheme which estimates the channel state in a simple manner, and adaptively switches its operation mode. It provides higher throughput than other comparable ARQ schemes in conditions of Land-Mobile-Satellite (LMS) channel.

In this contribution we present transform and neural network approaches to the interpolation of images. From transform point of view, the principles from [1] are modified for 1st and 2nd order interpolation. We present several new interpolation discrete orthogonal transforms. From neural network point of view, we present interpolation possibilities of multilayer perceptrons. We use various configurations of neural networks for 1st and 2nd order interpolation. The results are compared by means of tables.