We present a general approach to the development of adaptive image processing operators of variable shape and size that can be applied directly to non-uniformly sampled images. The approach is demonstrated for second order derivative operators using content-based compressed images.
Practical situations arise in which image data may be sparse and irregular in location: through incomplete sensing or compression techniques. Subsequent image processing usually requires image reconstruction/interpolation, which is computationally expensive and approximate. This paper pioneers direct processing of incomplete/compressed image data without image reconstruction. The work builds on output in IEEE ICIP 2002 and 2003; has been extended in IEEE ICIP 2005 and 2006; has led to exchange of research resources (algorithm code) with one of the pioneers of mesh modeling compression techniques (Professor Yang, Illinois Institute of Technology); and is being developed in the EPSRC-funded project EP/C006283/1.