Authors : Sanjay M. Malode and M. V. Sarode
Page Nos : 63-69
Description :
Digital videos may be degraded by several spatial and temporal corrupting factors which include but are not limited to noise, blurring, ringing, blocking, flickering, and other acquisition, compression or transmission artifacts. In view of the joint presence of random and fixed-pattern noise (FPN), the FPN typically arises in raw images acquired by focal plane arrays (FPA), such as CMOS sensors or thermal micro bolometers, where spatial and temporal non uniformities in the response of each photo detector generate a pattern superimposed on the image approximately constant in time. The spatial correlation characterizing the noise corrupting the data acquired by such sensors invalidates the classic additive white Gaussian noise (AWGN) assumptions of independent and identically distributed and hence white– noise. The FPN removal task is prominent in the context of long wave infrared (LWIR) thermography and hyper spectral imaging. Existing denoising methods can be classified into reference-based (also known as calibration-based) or scene based approaches. Reference-based approaches first calibrate the FPA using (at least) two homogeneous infrared targets, having different and known temperatures, and then linearly estimate the non-uniformities of the data. However, since the FPN slowly drifts in time, the normal operations of the camera need to be periodically interrupted to update the estimate which has become obsolete. Differently, scene-based approaches are able compensate the noise directly from the acquired data, by modeling the statistical nature of the FPN. This survey paper elaborates various approaches for noise removal and advancements in the image.
