Spatial Shape Error Concealment for Object-Based Image and Video Coding

Abstract

In this paper, an original spatial shape error-concealment technique, to be used in the context of object-based image and video coding schemes, is proposed. In this technique, it is assumed that the shape of the corrupted object at hand is in the form of a binary alpha plane, in which some of the shape data is missing due to channel errors. From this alpha plane, a contour corresponding to the border of the object can be extracted. However, due to errors, some parts of the contour will be missing and, therefore, the contour will be broken. The proposed technique relies on the interpolation of the missing contours with Bézier curves, which is done based on the available surrounding contours. After all the missing parts of the contour have been interpolated, the concealed alpha plane can be easily reconstructed from the fully recovered contour and used instead of the erroneous one improving the final subjective impact.

Bezier curve-based generic shape encoder

“…In particular, their robustness in curve and surface representation means they pervade many fields of multimedia technology including shape description of characters [1,2] and objects [3], active shape lip modelling (ASLM) [4], shape error concealment for MPEG-4 objects [5] and surface mapping [6]. The classical BC is defined by a set of control points (CP) with the number and orientation of these points governing the overall size and shape of the curve.…”

Image Dependent Spatial Shape Error Concealment for Multiple Shapes

“…These techniques have the advantage of having access to past information and therefore can perform well in a video sequence where an object's shape changes very little between consecutive frames. In contrast, when shape information changes significantly, including the emergence of new objects and object occlusion between frames, temporal methods alone are inadequate [13] and also obviously they are not the most appropriate for still images. In these circumstances spatial techniques are preferable, where the strong neighbouring interpixel spatial correlation is exploited to conceal erroneous pixels using information from correctly received and previously concealed pixels within a frame [11].…”

“…This exploits spatial redundant information on a statistical basis, though it does not use shape as a salient feature and so has subsequently been outperformed by those approaches which incorporate shape characteristics [11]. They employ parametric curves to geometrically conceal a lost boundary from correctly decoded shape information, with Bezier curves [4,13] and Hermite splines [11] used to conceal errors, though neither has information available upon the lost parts of the contour and so depends upon just decoded shape data. Since the control points for these curves are calculated from the tangents at the two contour-ends associated with each lost segment, their performance is highly dependent on the respective tangent vectors.…”

“…The aforementioned shape error concealment techniques [4,[11][12][13]15] only consider a single closed shape at a time. Consequently, application scenarios with multiple shapes (both closed and open ended) for example, palmprint representations, are not addressed.…”

Quasi-Bezier curves integrating localised information

“…While their origin can be traced back to the design of car body shapes in the automobile industries, their usage is no longer confined to this field. Indeed, their robustness in curve representation means BC now pervades many areas of multimedia technology, including shape description of characters [1] and objects [2], shape coding and error concealment for MPEG-4 coded objects [3].…”