Segmentation improvement of pig contour based on registration and fusion of IR thermal and optical images

To improve the effect of pig contour segmentation, an image preprocessing method is proposed based on registration and fusion of infrared (IR) thermal and optical images. An infrared thermal imager is used to acquire the thermal image and optical image of pig at the same time. The weighted partial Hausdorff distance (WPHD) is used as the similarity measurement to register the two source images. A resilient propagation (RPROP) algorithm is adopted to accelerate the iteration process. After decomposing the registered images by non-subsample contourlet transform (NSCT) separately, a fusion rule for low-pass sub-band coefficients is proposed by weighted averaging the combined values of the two images. This combined value is constructed by the average energy and variance of neighbor area. For the fusion of band-pass sub-band coefficients, the rule based neighbor area energy maximum is adopted. Then a fusion image is reconstructed through inverse transform of NSCT. Experimental results show that the fusion image has better visual effect and the segmental result of pig contour is more accurate and robust.