According to (Singh 1989), Change detection ( CD ) is “the process of identifying differences in the state of an object or phenomenon by observing it at different times”. Change detection from remote sensing images is very important in Geomatics’ applications such as urban and rural management, environmental applications, forest monitoring, agriculture, and so on. There are some essential processing tasks including geometric and radiometric corrections, feature extraction and selection, and classification, which must be performed on multi-temporal images, to do change detection. Radiometric correction is one of the most important tasks in change detection, since atmosphere condition, seasonal effects, and the change on view angle of sensors, and so on make a difference on illuminations of a pixel in two dates. There are many errors in the outcome of change detection without applying radiometric correction task. For this reason, a radiometric correction approach was developed in this study to do an appropriate change detection method. There are some radiometric correction approaches, which works based corresponding pixels. They use linear regression, neural network, and so on to make a relation among digital values of images in time series images. Undoubtedly, the corresponding pixels play a key role in radiometric correction approaches to solve unknown parameters of linear regression and neural network. Then, we proposed an efficient approach to select proper corresponding pixels.
In this research, three radiometric correction approaches including Pseudo Invariant Features ( PIF ), Radiometric control sets method ( RCSM ), and linear regression along with a new radiometric correction approach were implemented. To evaluate radiometric correction approaches, two case studies including Lake Urmia and Alaska were chosen. Landsat TM satellite images were also obtained from the study areas. Results show that the accuracy of change detection outcomes depends on threshold values of radiometric correction approaches. Moreover, overall accuracies of linear regression, pseudo invariant features, radiometric control sets method, and our proposed approach in the first and second study areas were (68 % , 93 % , 91 % , 94 % ) and (92 % , 95 % , 97 % , 97 % ), respectively. It indicates that our proposed radiometric correction approach is more efficient than linear regression, Pseudo Invariant Features, and radiometric control sets method. In the future woks, it is recommended to use non-linear functions instead of linear one in our proposed approach to achieve more accurate results.