Enhanced SRCNN for red culture image super-resolution reconstruction

Red culture images were taken in an earlier period, and the limited level of early camera shooting technology made the clarity of the pictures taken insufficient, which led to uncertainty in the corresponding historical research and enhanced the difficulty of subsequent image identification and retrieval. This paper aims to leverage image super-resolution reconstruction technology to enhance the information representation capabilities of red culture images. To this end, we introduce a super-resolution reconstruction dataset specifically tailored for the realm of red culture, accompanied by a lightweight deep learning model known as Enhanced SRCNN (ESRCNN). This model is proficient in processing both color and grayscale images. To address issues such as limited feature extraction in the SRCNN model, loss of shallow details, and ineffective utilization of feature information, the VGG-Skip-Channel (VSC) feature extraction module is introduced. Finally, the MS image reconstruction module is introduced to address the issue of local feature information loss, which arises from the use of large convolution kernels in image reconstruction by the SRCNN model. The results indicate that ESRCNN not only reduces the number of parameters but also achieves performance levels comparable to several widely utilized deep learning models for super-resolution reconstruction.