This paper proposes image inpainting algorithms in which the pixel values of the regions to be inpainting are recursively estimated by using multi-layered Perceptron from the original input image and its wavelet transformation. Instead of forward estimation by using the deep neural network such as convolutional neural network (CNN), we use shallow neural network and the pixel values are recursively estimated. To improve the estimation quality, wavelet transformation is also used as the input of the neural network. The effectiveness of the proposed approach was experimentally confirmed by using the face databases.
CITATION STYLE
Fujishige, H., Miyao, J., & Kurita, T. (2017). Image inpainting by recursive estimation using neural network and wavelet transformation. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10639 LNCS, pp. 652–661). Springer Verlag. https://doi.org/10.1007/978-3-319-70136-3_69
Mendeley helps you to discover research relevant for your work.