Content-Oriented Learned Image Compression

AbstractIn recent years, with the development of deep neural networks, end-to-end optimized image compression has made significant progress and exceeded classic methods in terms rate-distortion performance. However, most learning-based are unlabeled do not consider semantics or content when optimizing model. In fact, human eyes have different sensitivities to content, so also needs be considered. this paper, we propose a content-oriented method, which handles kinds contents strategies. Extensive experiments show that proposed method achieves competitive subjective results compared state-of-the-art learned methods.KeywordsImage compressionContent-orientedLoss metric