An Automated Multi-temporal Image Co-registration Technique for Lake-rich Environments Based on Pseudo Invariant Features

The increasing availability of remotely sensed data has facilitated an efficient and in-depth study of lake dynamics in the context of global change. Regional-scale lake monitoring requires effective change detection using multi- temporal and multi-sensor satellite imagery. Change detection involves a pixel-by-pixel comparison of multi- temporal images, necessitating precise image co-registration. The performance of widely-used co-registration techniques that rely upon manual selection of tie points is limited by the accuracy of individual tie points identified from the image pair to be co-registered. When the number of image pairs to be co-registered is large, these interactive methods however become prohibitive due to the significant volume of data. Therefore, the use of automated image co-registration techniques is required. The performance of area-based co-registration techniques is limited by factors such as atmospheric degradations, illumination effects, and sensor sensitivity variations in multi-temporal images, whereas feature- based techniques are robust to the above limiting factors. Feature based co-registration techniques establish correspondence between the features derived from the respective images, eliminating the need for radiometric normalization. However, the performance of such techniques is limited in lake-rich areas such as Arctic, where lakes are the dominant surface features and change over time. To overcome this limitation, we introduce the concept of Pseudo Invariant Features (PIFs) based on feature shape criteria to identify stable lakes - lakes that have not undergone significant change in shape over the time, and propose an approach that employs their center points in developing the co-registration model. Performance of the proposed approach is evaluated quantitatively, and sub-pixel co-registration accuracy is achieved.