Quality Assessment Techniques for Small Satellite Images

Abstract

Recently, the occupancy of small satellite in space is increasing, and one potential application of this is satellite imaging. The difference in acquisition conditions of satellite images makes it susceptible to much degradation compared to normal images. The common degradations are motion blur, low resolution and noise due to low light conditions, rain, snow, etc. Since small satellites are planned to be launched in big numbers, the number of image data to be analyzed will be higher, and this calls for some automatic techniques for assessing the quality of acquired images. Even though perceptual image quality assessment is a much explored area, there is not much work done in the area of satellite image quality assessment. In this paper, we review different techniques used for satellite image quality assessment and suggest some techniques which can also be used. The first method that we propose extracts GIST feature from satellite images and then uses a shallow neural network to predict the quality of the image from these features. The second method uses a deep convolutional neural network (CNN) with a shallow fully connected neural network to predict satellite image quality.