An EM-CI based approach to fusion of IR and visual images

With the cost decline of infrared (IR) cameras, it is envisage that more IR camera will be deployed in vision surveillance system for around-clock day and night surveillance. Infrared camera sense radiation emitted by an object at a non-zero absolute temperature in the infrared spectrum which is not available in visual image, but lose information, such as texture, color and geometric, which is available in visual cameras. Fusion of IR and visual images can enhance features in both kinds of images. And more impressively, it can reveal some new features that might not be present either in IR images or in visual images. In this paper, a statistical signal processing approach based on expectation maximization (EM) is proposed for IR and visual image fusion. The sensor images are described as the true scene corrupted by additive Gaussian distortion. At each iteration of the EM, the fusion result is obtained by using covariance intersection (CI) in the E-step, while the model parameters are updated in the M-step. The simulation results using the real IR and visual images demonstrate the effectiveness of the proposed method.