Defects like bubble and solid inclusion may disrupt the photovoltaic manufacturing process. The reliability to detect the defect of Photovoltaic Glass is particularly important. In order to separate defects from periodic background, this study introduces a novel approach to estimate saliency using image contrast and image signature. On the one hand, candidate saliency is computed based on center-surround contrast. On the other hand, original image is firstly converted by DCT transform, then the sign operation is applied to produce the image signature and the reconstructed image is computed by IDCT transform. Later the Gaussian kernel is applied to gain the reconstructed saliency. Finally, the reconstructed saliency is used to fuse the candidate saliency. Experiment results show that the proposed method gains relatively accurate saliency regions compared to the 7 methods.