Steganography, the practice of concealing hidden messages in digital media, has advanced significantly with the adoption of deep learning techniques. In this paper, we propose an advanced image steganography framework that leverages deep learning and attention mechanisms to achieve robust, imperceptible data hiding. The framework integrates Convolutional Neural Networks (CNNs) with Squeeze-and-Excitation (SE) blocks and Multi-Head Self-Attention layers to enhance feature extraction and selection, minimizing distortions in both cover and stego-images. The encoder-decoder architecture simultaneously trains hiding and revealing networks to ensure accurate embedding and recovery of the secret image. It leverages convolutional layers with varying kernel sizes, SE blocks for adaptive feature recalibration, and multi-head attention mechanisms to identify optimal embedding regions. Gaussian noise is added in the decoder to enhance robustness against distortions, while a custom loss function, combining reconstruction and distortion losses, optimizes the training process. Our experimental evaluations demonstrate superior performance compared to traditional LSB-based and deep learning methods, with high Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) values confirming image fidelity and imperceptibility. Our framework exhibits strong generalization across diverse datasets and payload capacities, representing a significant advancement in deep learning-based image steganography for secure data communication and watermarking.