CNN-BASED UNSUPERVISED REGISTRATION OF TIME-LAPSE MICROSCOPY IMAGE SEQUENCES

Abstract. Image registration is widely used in live cell microscopy image analysis to compensate for the motion. It a challenging task as not only moving (which causes rotation and translation), but also changes its form time making motion non-rigid. To address this, we propose CNN-based unsupervised method non-rigid of sequences. Our network predicts both deformation field between pair images sequence an affine transformation matrix compensation. The can be alone or combination with other approaches. proposed approach was successfully applied real sequences.We conducted experimental comparison existing methods including contour-based, intensity-based deep learning based joint denoising method. In addition, analyzed different regularizers their impact on alignment results. contour-based outperformed approaches average accuracy two metrics standard evaluation dataset.