High dynamic range image sensor using tone mapping operation

TA Engineering (General). Civil engineering (General)

Mughal, Waqas ; Engineering & Physical Sciences Research Council (EPSRC) ; Wasige, Edward

University:University of Glasgow

Department:School of Engineering

关键词: CMOS image sensor, HDR image sensors, tone mapping, fixed pattern noise.;   

Others  :  http://theses.gla.ac.uk/8151/1/2017MughalPhD.pdf

来源: University of Glasgow

【 摘 要 】

In nature, the dynamic range of a scene can be on the order of 1000000 to 0 from bright to dark areas. The human eye is capable of capturing images over a wide dynamic range of illumination while CMOS image sensors have limited ability to capture dynamic range available in nature. A Wide dynamic range (WDR) image is captured and displayed on low dynamic range (LDR) conventional media by a technique called tone mapping (TM). It is reported that most of the tone mapping operators have a monotonically increasing transduction function. A new CMOS pixel is proposed that aims to capture and display the wide dynamic range of illuminations by using the tone mapping monotonically increasing reference function. Different tone mapping functions like Reinhard photographic operator, Drago operator, etc. reference function are proposed and discussed. The tone mapping monotonically increasing function enables new pixel to capture WDR up to 6 decades of intensity.A pixel model of different tone mapping operators is developed with parameters that characterise the pixel response. It is proposed that the pixel model of certain tone mapping operators exists. In addition, the parameters extraction routine is also discussed. The pixel model response is compared with the simulated response to check its validity. The model error is presented and suggests that the developed model fits well with the simulation data. Fixed pattern noise (FPN) limits the performance of image sensors, and it is mainly due the variations between the responses of individual pixels within an array of pixels. A simple procedure has been adapted to reduce FPN in which parametric response of the pixel is used, with FPN modeled as variations in the individual parameters. The parameters of each pixel are measured, recorded and then used to correct their response.