Real-time object detection or recognition technology becomes more important for various intelligent vision systems. Processing models for object detection or recognition from natural images should tolerate pattern deformations and pattern position shifts. The hierarchical convolutional neural networks are considered as a promising model for robust object detection/recognition. This model requires huge computational power for a large number of multiply-and-accumulation operations. In order to apply this model to robot vision or various intelligent real-time vision systems, its LSI implementation is essential. This paper proposes a new algorithm for reducing multiply-and-accumulation operation by sorting neuron outputs by magnitude. We also propose an LSI architecture based on this algorithm. As a proof of concept for our LSI architecture, we have designed, fabricated and tested two test LSIs: a sorting LSI and an image-filtering LSI. The sorting LSI is designed based on the content addressable memory (CAM) circuit technology. The image-filtering LSI is designed for parallel processing by analog circuit array based on the merged/mixed analog-digital approach. We have verified the validity of our LSI architecture by measuring the LSIs.
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Osamu NOMURA, Takashi MORIE, Keisuke KOREKADO, Teppei NAKANO, Masakazu MATSUGU, Atsushi IWATA, "An Image-Filtering LSI Processor Architecture for Face/Object Recognition Using a Sorted Projection-Field Model Based on a Merged/Mixed Analog-Digital Architecture" in IEICE TRANSACTIONS on Electronics,
vol. E89-C, no. 6, pp. 781-791, June 2006, doi: 10.1093/ietele/e89-c.6.781.
Abstract: Real-time object detection or recognition technology becomes more important for various intelligent vision systems. Processing models for object detection or recognition from natural images should tolerate pattern deformations and pattern position shifts. The hierarchical convolutional neural networks are considered as a promising model for robust object detection/recognition. This model requires huge computational power for a large number of multiply-and-accumulation operations. In order to apply this model to robot vision or various intelligent real-time vision systems, its LSI implementation is essential. This paper proposes a new algorithm for reducing multiply-and-accumulation operation by sorting neuron outputs by magnitude. We also propose an LSI architecture based on this algorithm. As a proof of concept for our LSI architecture, we have designed, fabricated and tested two test LSIs: a sorting LSI and an image-filtering LSI. The sorting LSI is designed based on the content addressable memory (CAM) circuit technology. The image-filtering LSI is designed for parallel processing by analog circuit array based on the merged/mixed analog-digital approach. We have verified the validity of our LSI architecture by measuring the LSIs.
URL: https://global.ieice.org/en_transactions/electronics/10.1093/ietele/e89-c.6.781/_p
Copy
@ARTICLE{e89-c_6_781,
author={Osamu NOMURA, Takashi MORIE, Keisuke KOREKADO, Teppei NAKANO, Masakazu MATSUGU, Atsushi IWATA, },
journal={IEICE TRANSACTIONS on Electronics},
title={An Image-Filtering LSI Processor Architecture for Face/Object Recognition Using a Sorted Projection-Field Model Based on a Merged/Mixed Analog-Digital Architecture},
year={2006},
volume={E89-C},
number={6},
pages={781-791},
abstract={Real-time object detection or recognition technology becomes more important for various intelligent vision systems. Processing models for object detection or recognition from natural images should tolerate pattern deformations and pattern position shifts. The hierarchical convolutional neural networks are considered as a promising model for robust object detection/recognition. This model requires huge computational power for a large number of multiply-and-accumulation operations. In order to apply this model to robot vision or various intelligent real-time vision systems, its LSI implementation is essential. This paper proposes a new algorithm for reducing multiply-and-accumulation operation by sorting neuron outputs by magnitude. We also propose an LSI architecture based on this algorithm. As a proof of concept for our LSI architecture, we have designed, fabricated and tested two test LSIs: a sorting LSI and an image-filtering LSI. The sorting LSI is designed based on the content addressable memory (CAM) circuit technology. The image-filtering LSI is designed for parallel processing by analog circuit array based on the merged/mixed analog-digital approach. We have verified the validity of our LSI architecture by measuring the LSIs.},
keywords={},
doi={10.1093/ietele/e89-c.6.781},
ISSN={1745-1353},
month={June},}
Copy
TY - JOUR
TI - An Image-Filtering LSI Processor Architecture for Face/Object Recognition Using a Sorted Projection-Field Model Based on a Merged/Mixed Analog-Digital Architecture
T2 - IEICE TRANSACTIONS on Electronics
SP - 781
EP - 791
AU - Osamu NOMURA
AU - Takashi MORIE
AU - Keisuke KOREKADO
AU - Teppei NAKANO
AU - Masakazu MATSUGU
AU - Atsushi IWATA
PY - 2006
DO - 10.1093/ietele/e89-c.6.781
JO - IEICE TRANSACTIONS on Electronics
SN - 1745-1353
VL - E89-C
IS - 6
JA - IEICE TRANSACTIONS on Electronics
Y1 - June 2006
AB - Real-time object detection or recognition technology becomes more important for various intelligent vision systems. Processing models for object detection or recognition from natural images should tolerate pattern deformations and pattern position shifts. The hierarchical convolutional neural networks are considered as a promising model for robust object detection/recognition. This model requires huge computational power for a large number of multiply-and-accumulation operations. In order to apply this model to robot vision or various intelligent real-time vision systems, its LSI implementation is essential. This paper proposes a new algorithm for reducing multiply-and-accumulation operation by sorting neuron outputs by magnitude. We also propose an LSI architecture based on this algorithm. As a proof of concept for our LSI architecture, we have designed, fabricated and tested two test LSIs: a sorting LSI and an image-filtering LSI. The sorting LSI is designed based on the content addressable memory (CAM) circuit technology. The image-filtering LSI is designed for parallel processing by analog circuit array based on the merged/mixed analog-digital approach. We have verified the validity of our LSI architecture by measuring the LSIs.
ER -