Eyeblinks are having the significance to analyze the attention, fatigue, behaviour and emotion of an individual. Eyeblink recognition is adopted by many medical and surveillance applications to identify the person's state. The eye blink recognition on videos requires tracking the eye region and to count the number of eye blinks. In this paper, a three-stage model is presented to detect the eye blinks accurately. In the first stage, the frame similarity analysis, background separation, positional and mathematical filters are applied collectively to identify the effective eye region on unique frames. In the second stage, the similarity analysis using wavelet decomposition and statistical filters are applied on the segmented eye region. The filtered evaluation is performed to identify the change on the eye region of continuous segmented frames. At the final stage, distance driven map on structural and statistical features is applied to remove the invalid frame changes and to obtain the accurate eye blink count. The proposed model is applied on real time, web-collected and the NRC-IIT dataset videos. These complex videos are associated to the indoor and outdoor environments. The news reading and other complex video sequences are analyzed in this research. The observations identified that the proposed model has reduced the possible generated errors and provided the accurate detection of eye blinks.