Assessment of Quality for Operationally Generated Meteorological Satellite Image Products

Abstract

This thesis aims towards automatic quality assessment of meteorological satellite images captured by IMAGER payload of Indian National Satellite-3D (INSAT-3D). INSAT-3D is an advanced weather satellite of an India capable of generating images of earth disk in six different wavelengths. Aim is achieved by developing Image Quality Metrics (IQMs) that assigns score to an image based on details present. More specifically, this thesis focuses on measuring quality of meteorological satellite images without considering any reference image as reference image may not be assured all time. Also, desirable IQMs’s result should be consistent with the human observer judgments. Out of various image attributes that affect perceived quality of an image, only sharpness, contrast and noise attributes are considered only. This thesis explores various No Reference (NR) IQMs for sharpness and contrast estimation of an image, tested all of them against four types of distortion namely blur, contrast, sharpness and noise for all six channel images of IMAGER payload. Their applicability for sharpness, blur and contrast attributes is observed. For each channel, one sharpness and contrast measure is selected after behavior studied. Various noise estimation methods are surveyed. Noise estimation method is finalized out of all surveyed which is suitable for both low level and high level noisy images. All measures are implemented using APIs provided by Orfeo Toolbox (OTB) which provides libraries for standard remote sensing operations. Diurnal variation in terms of contrast, sharpness and noise attributes is computed and analyzed. Noise estimation range is observed based on profile generated by diurnal variation for each channel of IMAGER payload. In order to generate one overall index that assigns score to an image by considering impact of sharpness, contrast and noise together in overall quality perception, weight factors are selected for sharpness, contrast and noise attributes of an image for each six channels. The overall quality index is then computed as weighted sum of these three measures with the particular values of the respective weight obtained from subjective quality tests.