EE Seminar: Yehuda Odes

~~Yehuda Odes, 
M.Sc. student under the supervision of Prof. Shlomo Weiss and Dr. Hedva Spitzer

Wednesday, February 11th, 2015 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Algorithm for Adaptively Correcting Intensity Flickering in Video

Abstract
The 'constant image brightness' (CIB) assumption is of paramount importance in many computer vision algorithms, including, but not limited to, motion estimation and object recognition in video applications. Reducing large intensity flickering is also beneficial for human operators viewing the video. While prevalent in old video footage and time lapse sequences, modern high quality video sequences may also exhibit large intensity differences. This occurs largely due to environmental changes and as a result of various image enhancement algorithms applied in post processing.
We introduce a new framework for adaptively correcting flickering in videos. Taking cue from visual psychophysics (the relationship between physical stimuli and the perceptions they affect), we treat the various plausible factors determining the influence of the flickering. The factors include the size, intensity, temporal history and other parameters defining the flicker. By taking into account these free parameters we are able to adaptively correct intensity flickers in a more general and adaptive manner than was previously possible. The algorithm was successfully tested on approximately 20 real video sequences including time-lapse, Hyperlapse and surveillance camera footage.
Previous approaches to fix flickering are mainly based on histogram matching techniques, iterative solutions, and setting custom-tailored equations to mitigate a specific type of intensity flicker. These methods yield adequate results under certain conditions. However, more complex videos such as sequences captured by non-stationary cameras with content displaying a combination of global and local flickers, object occlusions, combined with scene illumination changes do not appear to be successfully de-flickered using the current techniques.
This research has been done as part of an Israeli government Magneton program.