A Low-Cost Hardware Design of Learning-Based One-Dimensional Interpolation for Real-Time Video Applications at the Edge

Shiau, Yeu-Horng; Huang, Kuo-Feng; Chen, Pei-Yin; Kuo, Cho-Yu

doi:10.1109/jetcas.2021.3121070

Cited by 1 publication

(11 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared to the hardware implementation of the CNN-based method [25]- [28], our LCDI-7+ only required much fewer hardware resources without much loss of the quality of the reconstructed image. Note that the comparison results with [27] are not provided in [24], but we get the same conclusion mentioned above.…”

Section: Introductionsupporting

confidence: 71%

“…As mentioned in the introduction above, SI [20] and LCDI [24] are the base related works in this paper, and they can be regarded as 2-D and 1-D interpolations, respectively. The main advantage of 1-D interpolation architecture is the reduction of computational complexity compared to the 2-D interpolation architecture.…”

Section: Related Workmentioning

confidence: 99%

“…Since the hardware resource is limited, some studies have developed novel interpolation-based SR methods [19]- [24], combining machine learning instead of the CNN-based SR method. They employ interpolation architecture and exploit pre-learned weights stored in a look-up table to calculate interpolated pixels instead of the polynomial function.…”

Section: Introductionmentioning

confidence: 99%

“…Likewise inspired by SI, we observe that it is still ameliorable; we proposed a modified method of SI in our previous work [24], called Learning-based Cross-directional Interpolation (LCDI). The main difference between SI and LCDI is the dimension of the referenced patches.…”

Section: Introductionmentioning

confidence: 99%

“…Kim et al [28] designed a novel CNN-based SR model and its hardware: Real-time SRCNN (RTSRCNN). In [24], we proposed an efficient SR method and its hardware implementation, but we only developed x2 SR architecture, and there is a problem that the quantity of prelearned weights is too much. Therefore, in this work, we develop x3 and x4 SR architecture based on the LCDI-7+ and attempt to reduce the required pre-trained weights.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A Cost-Effective Interpolation for Multi-Magnification Super-Resolution

2022

Self Cite

View full text Add to dashboard Cite

Super-Resolution (SR) was an important research topic, and SR methods based on Convolutional Neural Network (CNN) confirmed its groundbreaking performance. However, notably implementing the CNN model into resource-limited hardware devices is a great challenge. Therefore, we present a hardware-friendly and low-cost interpolation for Multi-Magnification SR image reconstruction. We follow our previous work, which is a learning-based interpolation (LCDI) with a self-defined classifier of image texture, and extend its original x2 architecture to x3 and x4 architecture. Besides, the required pre-trained weights are reduced by the fusion scheme. Experimentally, the proposed method has only 75% lower pre-trained weights than LCDI. Compared to the related work OLM-SI (One linear learning mapping-SI), the run-time and quantity of pre-trained weights of the x2 proposed method are at least 90% lower. Compared to CNN-based SR methods, the proposed method loses a little lower performance, but the evaluation of computational cost is much lower. In conclusion, the proposed method is cost-effective and a practical solution for resource-limited hardware and device.

show abstract