Prior Guided Face Super Resolution Using Convolution LSTM

Mangulley N.; Tomar A.S.; Arya K.V.; Pomachagua Y.; Rodriguez C. · Lecture Notes in Networks and Systems (2023) · DOI: 10.1007/978-981-99-1912-3_19

Resumen

Face Super-Resolution (SR) is the process of restoring a high-resolution (HR) face image from a low-resolution (LR) input. Numerous deep learning-based face SR models utilized the facial prior to super-resolved the input images. They estimate facial prior from the LR input, which may be inaccurate and thus causes them to learn the incorrect facial details. Therefore, this paper presents a prior guided model in which deep features are retrieved by utilizing the Convolution LSTM layers to develop a high-quality super-resolved face image. Also, the coarse SR image is used to approximate the parsing map prior and concatenated with the coarse image, which results in improving the quality of the output HR face images. The proposed model attains better quantitative as well as qualitative scores as compared to existing SR models.

Instituciones: Universidad Nacional Federico Villarreal