With the rapid advancement of intelligent photography, image cropping has become a crucial technique for optimizing visual perception and enhancing aesthetic quality. In manual cropping workflows, element grouping is a pivotal cognitive bridge, seamlessly linking perceptual understanding with aesthetic reasoning. It fundamentally influences the visual narrative’s semantic coherence and the final composition’s formal harmony. However, most existing cropping algorithms fail to identify pivotal element groupings and cannot adapt cropping decisions based on group-level visual characteristics guided by aesthetic principles. Thus, we propose Maximum Edge Weight Clique Driven Adaptive Image Cropping (MEWC-Crop) to bridge this gap. This method first extracts instance-level visual features via Faster R-CNN and Vision Transformer (ViT). Then, it explores inter-element relationships along spatial and structural dimensions, enabling fine-grained modeling of complex visual dependencies. Next, we reformulate the element grouping task as a maximum edge weight clique (MEWC) problem, which is solved via integer linear programming (ILP) to accurately identify subgraphs with strong internal connectivity and structural compactness. Finally, guided jointly by the visual properties of selected groups and photographic theory, MEWC-Crop adaptively adjusts both aspect ratio and layout configuration to generate diverse and aesthetically pleasing cropping results. Extensive experiments demonstrate that MEWC-Crop offers superior flexibility and diversity in cropping outputs, substantially improving aesthetic quality while pre-serving critical visual content.