Removal of blobs from retinal videos using intra-video registration

Medical images and videos are used for surgical planning and augmentation, post surgical analysis, disease progression, and cure estimation. Retinal videos help ophthalmologists in diagnosis and analysis of the different eye impairments. Such videos contain the essential information, however, illuminating devices used for capturing fine details result in formation of blobs in the captured video. Resultantly, visibility of the eye is masked and obscures the observation of underlying area. This paper proposes a method of removing the blobs (artifacts) from retinal videos based on automatic video registration. Rigid registration model is used for intra-video (inter-frame) registration; normalized mutual information is used as image similarity measure which is optimized using Powell´s method. The registration results show very good alignment between the frames and blobs are removed by replacing the area under them from the other frames. The results are checked on six retinal videos and they seem very promising.