To address the confusion and low accuracy of salient objects in co-saliency detection for image groups with complex environments, we proposed a co-saliency detection model based on objectness and a multi-layer linear model. First, we calculated the inter-saliency values using the background guidance factor weighted by saliency prior and objectness probability. We then designed a local region feature to calculate the intra-saliency values. The zero, first, and second Hu moments of the image were used to integrate the two-stage saliency values. Finally, saliency subgraphs were adaptively fused using a multi-layer linear model. Experimental results reveal that the AP scores of the proposed algorithm are 87.80% on iCoseg datasets and 83.50% on the MSRC dataset. Results from the evaluation of other experimental indicators are also improved significantly. The detected salient objects are more accurate and the adaptability of the algorithm is enhanced.