Edge-Avoiding Wavelets and their Applications (Yissum)
Almost every application for manipulating and improving photographic images, such as noise reduction and detail enhancement, relies on scale separation.
Linear translation-invariant (LTI) filtering is the classic and efficient method for achieving scale separation; however the method generates visual artefacts such as halos.
Edge-preserving smoothing filters avoid these artefacts but are computationally expensive, they are time-consuming and often increase memory requirements.
The new wavelets accelerate the computation of the edge-preserving smoothing as well as various other popular applications (such as image colorization) at no memory cost.New family of second-generation wavelets constructed using robust data-prediction lifting schemes. The new wavelets make it possible to compute nonlinear data-dependent multi-scale edge-preserving image filtering and processing at computation times which are linear in the number of image pixels.Avoids halo artefacts in band-independent multi-scale processing with minimal computational cost and without taking any special precautions.
Can be used to compute edge-aware interpolation schemes very efficiently, enabling real-time implementation of various applications such as image colorization
Avoids the difficulties of solving large and poorly conditioned systems of equations
Significantly accelerates various computational photography applications, achieving multi-scale data-dependent filtering at running times typical of linear translation-invariant (LTI) filtering.
The software is ready for incorporation into any image processing toolkit or video processing system to provide significantly faster results at reduced cost. It has particular benefits for:
Edge-preserving smoothing of color images
Edge-aware noise removal
Dynamic-range compression
Scattered data interpolation for image colorization
US patent Granted: 8,295,632
10-2009-2271
以色列
