Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Jin, Kyong Hwan | - |
dc.date.accessioned | 2021-10-05T08:30:14Z | - |
dc.date.available | 2021-10-05T08:30:14Z | - |
dc.date.created | 2021-06-18 | - |
dc.date.issued | 2021-05 | - |
dc.identifier.issn | 1070-9908 | - |
dc.identifier.uri | http://hdl.handle.net/20.500.11750/15400 | - |
dc.description.abstract | We discover that a trainable convolution layer with a stride over 1 and kernel ≥ stride is identical to a trainable block transform. A block transform is performed when we use a convolution layer with a stride ≥ 2 and a kernel ≥ the stride. For instance, if we use the same widths, such as a 2 × 2 convolution kernel and stride-2, there are no overlaps between sliding windows, so this layer operates a block transform on the partitioned 2 × 2 blocks. A block transform reduces the computational complexity due to a stride ≥ 2. To keep the original size, we apply a transposed convolution (stride = kernel ≥ 2), an adjoint operator of a forward block transform. Based on this relationship, we propose a trainable multi-scale block transform for autoencoders. The proposed method has an encoder consisting of two sequential convolutions with stride-2, a 2× 2 kernel, and a decoder consisting of the encoder's two adjoint operators (transposed convolution). Clipping is used for nonlinear activations. Inspired by the zero-frequency element in the dictionary learning method, the proposed method uses DC values for residual learning. The proposed method shows high-resolution representations, whereas the stride-1 convolutional autoencoder with 3 × 3 kernels generates blurry images. © 1994-2012 IEEE. | - |
dc.language | English | - |
dc.publisher | Institute of Electrical and Electronics Engineers | - |
dc.title | Deep Block Transform for Autoencoders | - |
dc.type | Article | - |
dc.identifier.doi | 10.1109/LSP.2021.3082031 | - |
dc.identifier.scopusid | 2-s2.0-85107189040 | - |
dc.identifier.bibliographicCitation | IEEE Signal Processing Letters, v.28, pp.1016 - 1019 | - |
dc.description.isOpenAccess | FALSE | - |
dc.subject.keywordAuthor | autoencoder | - |
dc.subject.keywordAuthor | Block transform | - |
dc.subject.keywordAuthor | convolutional neural network | - |
dc.subject.keywordAuthor | image representation | - |
dc.subject.keywordPlus | Zero frequency | - |
dc.subject.keywordPlus | Learning systems | - |
dc.subject.keywordPlus | Convolution | - |
dc.subject.keywordPlus | Signal encoding | - |
dc.subject.keywordPlus | Adjoint operators | - |
dc.subject.keywordPlus | Block transforms | - |
dc.subject.keywordPlus | Blurry images | - |
dc.subject.keywordPlus | Convolution kernel | - |
dc.subject.keywordPlus | Dictionary learning | - |
dc.subject.keywordPlus | High resolution | - |
dc.subject.keywordPlus | Sliding Window | - |
dc.citation.endPage | 1019 | - |
dc.citation.startPage | 1016 | - |
dc.citation.title | IEEE Signal Processing Letters | - |
dc.citation.volume | 28 | - |
There are no files associated with this item.