An image compression coding framework based on adaptive sub-sampling and super-resolution reconstruction is designed for the Joint Photographic Experts Group(JPEG) standard. At the encoder side, a variety of Different Sampling Modes(DSM) and Quantization Modes(QM) are designed for the original image to be encoded. Then, the rate distortion optimization algorithm selects the optimal downsampling and quantization modes from various modes. Finally, the image to be encoded will be sampled and compressed by the standard JPEG compression under the selected optimal mode. In the decoder side, the super-resolution reconstruction algorithm based on convolutional neural network is utilized to reconstruct the decoded sub-sampled image. In addition, the proposed framework is also effective and feasible under the JPEG2000 compression standard. The experimental results show that compared with the mainstream coding and decoding standards and advanced encoding and decoding methods, the framework can effectively improve the rate distortion performance and obtain better visual effects.