A Swin Transformer-based image semantic communication scheme that integrated channel state information (CSI) awareness with generative adversarial network (GAN) assistance was proposed aiming at the issues of texture loss and edge blurring in reconstructed images encountered by existing deep joint source-channel coding schemes for image semantic communication in order to improve the perceptual quality and robustness of image transmission under typical SNR conditions. A CSI-guided self-attention mechanism was designed at the encoder. A scalar CSI was mapped into an additive bias matrix through a multilayer perceptron and incorporated into the self-attention computation in order to dynamically reshape the feature extraction topology. A CSI-gated generative restoration module was developed at the decoder. CSI was nonlinearly mapped to generate a gating factor for dynamically adjusting the restoration weight of the GAN on residual features. The experimental results demonstrate that the proposed scheme achieves superior peak signal-to-noise ratio (PSNR) and multi-scale structural similarity (MS-SSIM) performance across datasets with different resolutions compared with conventional separated source-channel coding and existing mainstream semantic communication schemes. The proposed method provides an effective solution for image semantic transmission in complex channel environments.