{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,24]],"date-time":"2026-06-24T02:46:22Z","timestamp":1782269182309,"version":"3.54.5"},"reference-count":38,"publisher":"Springer Science and Business Media LLC","issue":"3","license":[{"start":{"date-parts":[[2026,5,27]],"date-time":"2026-05-27T00:00:00Z","timestamp":1779840000000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,5,27]],"date-time":"2026-05-27T00:00:00Z","timestamp":1779840000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/501100004270","name":"Royal Institute of Technology","doi-asserted-by":"crossref","id":[{"id":"10.13039\/501100004270","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["J Math Imaging Vis"],"published-print":{"date-parts":[[2026,6]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    This paper presents the results of analyzing and modeling a set of 8 \u201cmaster key filters\u201d, which have been extracted by applying a clustering approach to the receptive fields learned in depthwise-separable deep networks based on the ConvNeXt architecture. For this purpose, we first compute spatial spread measures in terms of weighted mean values and weighted variances of the absolute values of the learned filters, which support the working hypotheses that: (i)\u00a0the learned filters can be modeled by separable filtering operations over the spatial domain, and that (ii)\u00a0the spatial offsets of the those learned filters that are non-centered are rather close to half a grid unit. Then, we model the clustered \u201cmaster key filters\u201d in terms of difference operators applied to a spatial smoothing operation in terms of the discrete analog of the Gaussian kernel, and demonstrate that the resulting idealized models of the receptive fields show good qualitative similarity to the learned filters. This modeling is performed in two different ways: (i)\u00a0using possibly different values of the scale parameters in the coordinate directions for each filter, and (ii)\u00a0using the same value of the scale parameter in both coordinate directions. Then, we perform the actual model fitting by either (i)\u00a0requiring spatial spread measures in terms of spatial variances of the absolute values of the receptive fields to be equal, or (ii)\u00a0minimizing the discrete\n                    <jats:inline-formula>\n                      <jats:alternatives>\n                        <jats:tex-math>$$l_1$$<\/jats:tex-math>\n                        <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\">\n                          <mml:msub>\n                            <mml:mi>l<\/mml:mi>\n                            <mml:mn>1<\/mml:mn>\n                          <\/mml:msub>\n                        <\/mml:math>\n                      <\/jats:alternatives>\n                    <\/jats:inline-formula>\n                    - or\n                    <jats:inline-formula>\n                      <jats:alternatives>\n                        <jats:tex-math>$$l_2$$<\/jats:tex-math>\n                        <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\">\n                          <mml:msub>\n                            <mml:mi>l<\/mml:mi>\n                            <mml:mn>2<\/mml:mn>\n                          <\/mml:msub>\n                        <\/mml:math>\n                      <\/jats:alternatives>\n                    <\/jats:inline-formula>\n                    -norms between the idealized receptive field models and the learned filters. Complementary experimental results then demonstrate that the idealized models of receptive fields have very good predictive properties for replacing the learned filters by idealized filters in depthwise-separable deep networks, thus showing that the learned filters in depthwise-separable deep networks can be well approximated by discrete scale-space filters. Notably, we show that, for a reduced version of the ConvNeXt architecture, using a set of only 8 discrete scale-space filters leads to almost as good accuracy as for the receptive fields trained from scratch on the ImageNet dataset.\n                  <\/jats:p>","DOI":"10.1007\/s10851-026-01290-0","type":"journal-article","created":{"date-parts":[[2026,5,27]],"date-time":"2026-05-27T14:39:15Z","timestamp":1779892755000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Modeling and Analysis of the 8 Filters from the \u201cMaster Key Filters Hypothesis\u201d for Depthwise-Separable Deep Networks in Relation to Idealized Receptive Fields Based on Scale-Space Theory"],"prefix":"10.1007","volume":"68","author":[{"given":"Tony","family":"Lindeberg","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zahra","family":"Babaiee","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Peyman M.","family":"Kiasari","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,5,27]]},"reference":[{"key":"1290_CR1","unstructured":"Abramowitz, M., Stegun, I.A. (eds.): Handbook of Mathematical Functions, 55th edn. Applied Mathematics Series. National Bureau of Standards, US Government Printing Office (1964). (Reprinted by Dover Publications)"},{"key":"1290_CR2","unstructured":"Babaiee, Z., Kiasari, P., Rus, D., Grosu, R.: We need far fewer unique filters than we thought. In NeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning, (2024a). https:\/\/openreview.net\/pdf?id=RV8oJTkqkj"},{"key":"1290_CR3","unstructured":"Babaiee, Z., Kiasari, P.M., Rus, D., Grosu, R.: Unveiling the unseen: identifiable clusters in trained depthwise convolutional kernels. In International Conference for Learning Representations (ICLR 2024), (2024b). preprint at arXiv:2401.14469"},{"key":"1290_CR4","doi-asserted-by":"crossref","unstructured":"Babaiee, Z., Kiasari, P., Rus, D., Grosu, R.: The master key filters hypothesis: deep filters are general. In AAAI Conference on Artificial Intelligence, (2025a). preprint at arXiv:2412.16751","DOI":"10.1609\/aaai.v39i2.32175"},{"key":"1290_CR5","unstructured":"Babaiee, Z., Kiasari, P., Rus, D., Grosu, R.: The quest for universal master key filters in DS-CNNs. Proc. Neural Information Processing Systems (NeurIPS 2025), (2025b). preprint at arXiv:2509.11711"},{"key":"1290_CR6","doi-asserted-by":"crossref","unstructured":"Basting, M., Bruintjes, R.-J., Wiedemer, T., K\u00fcmmerer, M., Bethge, M., van Gemert, J.: Scale learning in scale-equivariant convolutional networks. Proc. Computer Vision, Imaging and Computer Graphics Theory and Applications (VISAPP 2024), 567: 574, 2024","DOI":"10.5220\/0012379800003660"},{"key":"1290_CR7","doi-asserted-by":"crossref","unstructured":"Chollet, F.: Xception: Deep learning with depthwise separable convolutions. In Proc. Computer Vision and Pattern Recognition (CVPR 2017), pages 1800\u20131807, (2017)","DOI":"10.1109\/CVPR.2017.195"},{"key":"1290_CR8","unstructured":"Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, Z., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: transformers for image recognition at scale. In International Conference on Learning Representations (ICLR 2021), (2021)"},{"key":"1290_CR9","doi-asserted-by":"crossref","unstructured":"Gavilima-Pilataxi, H., Ibarra-Fiallo, J.: Multi-channel Gaussian derivative neural networks for crowd analysis. In Proc. International Conference on Pattern Recognition Systems (ICPRS 2023), pages 1\u20137, (2023)","DOI":"10.1109\/ICPRS58416.2023.10179046"},{"key":"1290_CR10","doi-asserted-by":"crossref","unstructured":"He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In Proc. Computer Vision and Pattern Recognition (CVPR 2016), pages 770\u2013778, (2016)","DOI":"10.1109\/CVPR.2016.90"},{"key":"1290_CR11","unstructured":"Howard, A.G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., Adam, H.: Mobilenets: efficient convolutional neural networks for mobile vision applications. arXiv:1704.04861, (2017)"},{"key":"1290_CR12","first-page":"368","volume":"26","author":"T Iijima","year":"1962","unstructured":"Iijima, T.: Basic theory on normalization of pattern (in case of typical one-dimensional pattern). Bull. Electrotech. Lab. 26, 368\u2013388 (1962). ((in Japanese))","journal-title":"Bull. Electrotech. Lab."},{"key":"1290_CR13","doi-asserted-by":"crossref","unstructured":"Jacobsen, J.-J., van Gemert, J., Lou, Z., Smeulders, A.W.M.: Structured receptive fields in CNNs. In Proc. Computer Vision and Pattern Recognition (CVPR 2016), pages 2610\u20132619, (2016)","DOI":"10.1109\/CVPR.2016.286"},{"issue":"5","key":"1290_CR14","doi-asserted-by":"publisher","first-page":"363","DOI":"10.1007\/BF00336961","volume":"50","author":"JJ Koenderink","year":"1984","unstructured":"Koenderink, J.J.: The structure of images. Biol. Cybern. 50(5), 363\u2013370 (1984)","journal-title":"Biol. Cybern."},{"issue":"6","key":"1290_CR15","doi-asserted-by":"publisher","first-page":"597","DOI":"10.1109\/34.141551","volume":"14","author":"JJ Koenderink","year":"1992","unstructured":"Koenderink, J.J., van Doorn, A.J.: Generic neighborhood operators. IEEE Trans. Pattern Anal. Mach. Intell. 14(6), 597\u2013605 (1992)","journal-title":"IEEE Trans. Pattern Anal. Mach. Intell."},{"issue":"11","key":"1290_CR16","doi-asserted-by":"publisher","first-page":"2278","DOI":"10.1109\/5.726791","volume":"86","author":"Y LeCun","year":"1998","unstructured":"LeCun, Y., Bottou, L., Bengio, Y., Haffner, P.: Gradient-based learning applied to document recognition. Proc. IEEE 86(11), 2278\u20132324 (1998)","journal-title":"Proc. IEEE"},{"issue":"3","key":"1290_CR17","doi-asserted-by":"publisher","first-page":"234","DOI":"10.1109\/34.49051","volume":"12","author":"T Lindeberg","year":"1990","unstructured":"Lindeberg, T.: Scale-space for discrete signals. IEEE Trans. Pattern Anal. Mach. Intell. 12(3), 234\u2013254 (1990)","journal-title":"IEEE Trans. Pattern Anal. Mach. Intell."},{"key":"1290_CR18","doi-asserted-by":"crossref","unstructured":"Lindeberg, T.: Scale-Space Theory in Computer Vision. Springer, (1993a)","DOI":"10.1007\/978-1-4757-6465-9"},{"issue":"4","key":"1290_CR19","doi-asserted-by":"publisher","first-page":"349","DOI":"10.1007\/BF01664794","volume":"3","author":"T Lindeberg","year":"1993","unstructured":"Lindeberg, T.: Discrete derivative approximations with scale-space properties: a basis for low-level feature extraction. J. Math. Imaging Vis. 3(4), 349\u2013376 (1993)","journal-title":"J. Math. Imaging Vis."},{"key":"1290_CR20","doi-asserted-by":"crossref","unstructured":"Lindeberg, T.: On the axiomatic foundations of linear scale-space. In J. Sporring, M. Nielsen, L. Florack, and P. Johansen, editors, Gaussian Scale-Space Theory: Proc. PhD School on Scale-Space Theory, pages 75\u201397, Copenhagen, Denmark, (1996). Springer","DOI":"10.1007\/978-94-015-8802-7_6"},{"issue":"1","key":"1290_CR21","doi-asserted-by":"publisher","first-page":"36","DOI":"10.1007\/s10851-010-0242-2","volume":"40","author":"T Lindeberg","year":"2011","unstructured":"Lindeberg, T.: Generalized Gaussian scale-space axiomatics comprising linear scale-space, affine scale-space and spatio-temporal scale-space. J. Math. Imaging Vis. 40(1), 36\u201381 (2011)","journal-title":"J. Math. Imaging Vis."},{"issue":"6","key":"1290_CR22","doi-asserted-by":"publisher","first-page":"589","DOI":"10.1007\/s00422-013-0569-z","volume":"107","author":"T Lindeberg","year":"2013","unstructured":"Lindeberg, T.: A computational theory of visual receptive fields. Biol. Cybern. 107(6), 589\u2013635 (2013)","journal-title":"Biol. Cybern."},{"issue":"1","key":"1290_CR23","doi-asserted-by":"publisher","DOI":"10.1016\/j.heliyon.2021.e05897","volume":"7","author":"T Lindeberg","year":"2021","unstructured":"Lindeberg, T.: Normative theory of visual receptive fields. Heliyon 7(1), e05897 (2021). https:\/\/doi.org\/10.1016\/j.heliyon.2021.e05897","journal-title":"Heliyon"},{"key":"1290_CR24","doi-asserted-by":"crossref","unstructured":"Lindeberg, T.: Scale-covariant and scale-invariant Gaussian derivative networks. In A. Elmoataz, A., Fadili, J., Qu\u00e9au, Y., Rabin, J., and Simon, L. (eds.), Proc. Scale Space and Variational Methods in Computer Vision (SSVM 2021), volume 12679 of Springer LNCS, pages 3\u201314, (2021b)","DOI":"10.1007\/978-3-030-75549-2_1"},{"issue":"3","key":"1290_CR25","doi-asserted-by":"publisher","first-page":"223","DOI":"10.1007\/s10851-021-01057-9","volume":"64","author":"T Lindeberg","year":"2022","unstructured":"Lindeberg, T.: Scale-covariant and scale-invariant Gaussian derivative networks. J. Math. Imaging Vis. 64(3), 223\u2013242 (2022)","journal-title":"J. Math. Imaging Vis."},{"issue":"1\u20132","key":"1290_CR26","doi-asserted-by":"publisher","first-page":"21","DOI":"10.1007\/s00422-022-00953-6","volume":"117","author":"T Lindeberg","year":"2023","unstructured":"Lindeberg, T.: A time-causal and time-recursive scale-covariant scale-space representation of temporal signals and past time. Biol. Cybern. 117(1\u20132), 21\u201359 (2023)","journal-title":"Biol. Cybern."},{"issue":"5","key":"1290_CR27","doi-asserted-by":"publisher","first-page":"759","DOI":"10.1007\/s10851-024-01196-9","volume":"66","author":"T Lindeberg","year":"2024","unstructured":"Lindeberg, T.: Discrete approximations of Gaussian smoothing and Gaussian derivatives. J. Math. Imaging Vis. 66(5), 759\u2013800 (2024)","journal-title":"J. Math. Imaging Vis."},{"key":"1290_CR28","doi-asserted-by":"publisher","DOI":"10.3389\/frsip.2024.1447841","author":"T Lindeberg","year":"2025","unstructured":"Lindeberg, T.: Approximation properties relative to continuous scale space for hybrid discretizations of Gaussian derivative operators. Front. Signal Process. (2025). https:\/\/doi.org\/10.3389\/frsip.2024.1447841","journal-title":"Front. Signal Process."},{"key":"1290_CR29","doi-asserted-by":"crossref","unstructured":"Liu, Z., Mao, H., Wu, C.-Y., Feichtenhofer, C., Darrell, T., Xie, S.: A ConvNet for the 2020s. In Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition, pages 11976\u201311986, (2022)","DOI":"10.1109\/CVPR52688.2022.01167"},{"key":"1290_CR30","doi-asserted-by":"crossref","unstructured":"Penaud-Polge, V., Velasco-Forero, S., Angulo, J.: Fully trainable Gaussian derivative convolutional layer. In International Conference on Image Processing (ICIP 2022), pages 2421\u20132425, (2022)","DOI":"10.1109\/ICIP46576.2022.9897734"},{"issue":"3","key":"1290_CR31","doi-asserted-by":"publisher","first-page":"29","DOI":"10.1007\/s10851-025-01245-x","volume":"67","author":"A Perzanowski","year":"2025","unstructured":"Perzanowski, A., Lindeberg, T.: Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations. J. Math. Imaging Vis. 67(3), 29 (2025)","journal-title":"J. Math. Imaging Vis."},{"key":"1290_CR32","doi-asserted-by":"publisher","first-page":"8342","DOI":"10.1109\/TIP.2021.3115001","volume":"30","author":"SL Pintea","year":"2021","unstructured":"Pintea, S.L., T\u00f6men, N., Goes, S.F., Loog, M., van Gemert, J.C.: Resolution learning in deep convolutional networks using scale-space theory. IEEE Trans. Image Process. 30, 8342\u20138353 (2021)","journal-title":"IEEE Trans. Image Process."},{"key":"1290_CR33","doi-asserted-by":"crossref","unstructured":"Saldanha, N., Pintea, S.L., van Gemert, J.C., Tomen, N.: Frequency learning for structured CNN filters with Gaussian fractional derivatives. arXiv:2111.06660, (2021)","DOI":"10.5244\/C.35.388"},{"key":"1290_CR34","unstructured":"Sangalli, M., Blusseau, S., Velasco-Forero, S., Angulo, J.: Scale equivariant U-net. In Proc. British Machine Vision Conference (BMVC 2022), page 763, (2022)"},{"issue":"1","key":"1290_CR35","doi-asserted-by":"publisher","first-page":"209","DOI":"10.1007\/s10851-022-01114-x","volume":"65","author":"BMN Smets","year":"2023","unstructured":"Smets, B.M.N., Portegies, J., Bekkers, E.J., Duits, R.: PDE-based group equivariant convolutional neural networks. J. Math. Imaging Vis. 65(1), 209\u2013239 (2023)","journal-title":"J. Math. Imaging Vis."},{"issue":"3","key":"1290_CR36","doi-asserted-by":"publisher","first-page":"237","DOI":"10.1023\/A:1008344623873","volume":"10","author":"J Weickert","year":"1999","unstructured":"Weickert, J., Ishikawa, S., Imiya, A.: Linear scale-space has first been proposed in Japan. J. Math. Imaging Vis. 10(3), 237\u2013252 (1999)","journal-title":"J. Math. Imaging Vis."},{"key":"1290_CR37","doi-asserted-by":"crossref","unstructured":"Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I.S., Xie, S.: ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders. In Computer Vision and Pattern Recognition (CVPR 2023), (2023)","DOI":"10.1109\/CVPR52729.2023.01548"},{"key":"1290_CR38","doi-asserted-by":"crossref","unstructured":"Yang, Y., Dasmahapatra, S., Mahmoodi, S.: Scale-equivariant UNet for histopathology image segmentation. Proc. Machine Learning Research 194, 130\u2013148, (2022),","DOI":"10.2139\/ssrn.4348782"}],"container-title":["Journal of Mathematical Imaging and Vision"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10851-026-01290-0.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/article\/10.1007\/s10851-026-01290-0","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10851-026-01290-0.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,24]],"date-time":"2026-06-24T01:49:08Z","timestamp":1782265748000},"score":1,"resource":{"primary":{"URL":"https:\/\/link.springer.com\/10.1007\/s10851-026-01290-0"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,27]]},"references-count":38,"journal-issue":{"issue":"3","published-print":{"date-parts":[[2026,6]]}},"alternative-id":["1290"],"URL":"https:\/\/doi.org\/10.1007\/s10851-026-01290-0","relation":{},"ISSN":["0924-9907","1573-7683"],"issn-type":[{"value":"0924-9907","type":"print"},{"value":"1573-7683","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,5,27]]},"assertion":[{"value":"16 September 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"9 March 2026","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"27 May 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"T.L. is a member of the editorial board of Journal of Mathematical Imaging and Vision.","order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Conflict of interest"}}],"article-number":"22"}}