Non-Linear Correlation Filter Based Image Recognition and Tracking

Objectives: The proposed method was to trace the image, even if you enable the reliable and accurate image tracking the movement of the image without calibration process by using a non-linear correlation filter large. Methods/Statistical analysis: By introducing the concept of the penalty associated with neighborhood function proposes an image similarity measure derived by non-linear transformation of the correlation filter. Picture image exhibited excellent results in the case that the motion picture track of a large set of experimental results faces average of 90%, average 85% for the other users. Findings: It proposed a new method for image tracking. The proposed system as compared to existing eye tracking system could track an image accurately and reliably even eliminating the need for calibration fluid motion. Improvements/ Applications: Scale image by tracing the method according to the changes and further applied to a method for removing a background element that does not move, and to develop faster and more accurate image tracking system.