Segmentation of transparent objects plays a crucial role in the field of computer vision, especially concerning scene comprehension and perception. Contrast to conventional segmentation, transparent objects are difficult to accurately segment due to blending, blurred boundaries, and susceptibility to light interference. Considering these factors, segmentation of transparent objects relies heavily on accurate edge positioning and global consistency. Simultaneously, it’s necessary to reduce the influence of lighting during the segmentation process. To solve these problems, we propose an efficient segmentation architecture for transparent objects known as edge enhanced and global integrated transformers (EG-Trans). Based on the linear sensitivity of edges, we first propose an edge enhanced multi-head self-attention mechanism, which efficiently extracts edge features. Additionally, we design a global compensation block that fully leverages local and global features to achieve global compensation for the transformer block. In the end, we introduce a multi-scale fusion block that benefits from feature channel interaction and multi-scale convolutional fusion, effectively mitigating the impact of lighting variations. Benchmark experiments demonstrate that our EG-Trans achieves superior performance compared to other state-of-the-art models on challenging public benchmark datasets. Even under challenging lighting variations, EG-Trans maintains high-quality segmentation. The codes are available at https://github.com/cehn-jiawei/EG-Trans .