Summary of the technology
We presented a new example-based image upscaling method that performs less nearest-patch computations and uses a custom designed ﬁlter banks to synthesize the image explicitly. The faster search is based on a local self-similarity observation that we point out in natural images, where edges and other singularities are locally scale invariant. The tests we performed to measure this invariance show that this assumption holds best for small scaling factors.
Comparisons show that the localized search, permitted by the local scale invariance, outperform approximate global searches both in quality and running time. We formulated and designed novel ﬁlter banks that allow us to perform such small, non-dyadic, image scalings. These ﬁlter banks extend the common dyadic transformations and are designed to model the image upscaling process. With these ﬁlters we achieve, using explicit computations, upscaled image which is highly consistent with the input image. Altogether, we propose a fully local high-quality upscaling algorithms that operates in real-time when implemented in a GPU.
Description of the technology
Description of the technology
New image processing algorithm based on example-based super-resolution that results in a high-quality and efficient single-image upscaling technique. Image upsampling exploits local scale-similarity in natural images. When downscaled, the patches (yellow squares) are very similar to their cropped version (red squares).
We propose a new high-quality and efﬁcient single-image upscaling technique that extends existing example-based super-resolution frameworks. In our approach we do not rely on an external example database or use the whole input image as a source for example patches. Instead, we follow a local self-similarity assumption on natural images and extract patches from extremely localized regions in the input image. This allows us to reduce considerably the nearest-patch search time without compromising quality in most images. Tests, that we perform and report, show that the local-self similarity assumption holds better for small scaling factors where there are more example patches of greater relevance. We implement these small scalings using dedicated novel non-dyadic ﬁlter banks, that we derive based on principles that model the upscaling process. Moreover, the new ﬁlters are nearly-biorthogonal and hence produce high-resolution images that are highly consistent with the input image without solving implicit back-projection equations. The local and explicit nature of our algorithm makes it simple, efﬁcient and allows a trivial parallel implementation on a GPU. We demonstrate the new method ability to produce high-quality resolution enhancement, its application to video sequences with no algorithmic modiﬁcation, and its efﬁciency to perform real-time enhancement of lowresolution video standard into recent high-deﬁnition formats.
Main advantages of its use
- Algorithm produces high-quality, sharp images
- Reduces the process search time without compromising quality
- Simple, efficient, and trivially parallel algorithm enables increase in the resolution of video
- sequences with no algorithmic modifications and at reasonable running times
- Up-scaling video and still images
About Yissum - Research Development Company of the Hebrew University
Technology Transfer Office from IsraelYissum - Research Development Company of the Hebrew University
Yissum Research Development Company of the Hebrew University of Jerusalem Ltd. Founded in 1964 to protect and commercialize the Hebrew University’s intellectual property. Ranked among the top technology transfer companies, Yissum has registered over 8,900 patents covering 2,500 inventions; has licensed out 800 technologies and has spun-off 90 companies. Products that are based on Hebrew University technologies and were commercialized by Yissum generate today over $2 Billion in annual sales.