In order to segment the helicopter target in the infrared image with complex cloud background effectively

a variational segmentation model based on manifold feature and shape priori is proposed. The region energy term is constructed on the basis of gray distribution of the image data to drive the deformation curve to target boundary. A regional covariance descriptor is introduced on Symmetric Positive Definite (SPD) matrix manifold to define a prior feature energy term to improve the robustness of the segmentation algorithm. The segmentation result of the helicopter in the first stage of the infrared image is obtained by the combination of the region term

the boundary term and the prior feature term. In the second stage

Principal Component Analysis(PCA) is used to capture the prior shape variation pattern of the infrared helicopter target

and the shape of evolution curve is reconstructed in PCA space as the prior knowledge to restrain the curve deforming. Finally

the complete contour of the infrared helicopter is obtained. Experimental results demonstrate that this method can effectively extract the whole contour of the infrared target.