Digital communication often uses non-verbal ways to transmit important information. Such as; images, symbols, and different textures. At the time of the wavelet transform, a specific block of image (a small section) is used to hide the secret information. Due to this, the secret image size needs to be changed before the embedding process, and this process also affects the extracted image quality. This paper proposes a secure image steganography technique based on the discrete wavelet transform (DWT) and deep learning (DL) to improve the quality of the stego image and the extracted secret image. In the embedding process initially cover image is transformed into DWT coefficients, then embed the scrambled secret image following by singular value decomposition (SVD) and alpha blending operation. To get the stego image, the inverse discrete wavelet transform (IDWT) is applied. The secret image extraction process is the inverse of the embedding process, but due to the wavelet transform, a compressed secret image is extracted. This secret image resolution is increased using, DL-based very deep super-resolution (VDSR) neural network in post-processing. It converts the extracted image according to the size required by the receiver. The proposed VDSR method is evaluated on a publicly available dataset, the IAPR TC-12 Benchmark (dataset link is given before reference section). The proposed method has a 51.66 to 38.69 dB peak signal-to-noise ratio (PSNR) and a 0.99 structural similarity index (SSIM) for the various alpha values, which is shown in Section 3.3. According to obtained results, there is a 99.9% similarity between the SSIMs of the original and attacked stego images that makes the proposed technique robust. The observed range of SSIM is from 99.9% to 100%.