Image sensor trained to classify optically projected images by reading the few most relevant pixels

Researchers are developing a sparse pixel image sensor, based on a two-dimensional array of metal-semiconductor-metal photodetectors with individually adjustable photosensitivity values.

New research paper “Sparse pixel image sensor” from Institute of Photonics at Vienna University of Technology.

“As conventional frame-based cameras suffer from high power consumption and latency, several new types of image sensors have been designed, some of them exploiting the scarcity of natural images in certain areas. transformation. Instead of sampling the complete image, these devices only capture the coefficients of the most relevant spatial frequencies. The number of samples can be even lower if a signal only needs to be classified rather than being entirely reconstructed. Based on the corresponding mathematical framework, we have developed an image sensor that can be trained to classify optically projected images by reading the few most relevant pixels. The device is based on a two-dimensional array of metal-semi photodetectors. -conductor-metal with individually adjustable photosensitivity values.We demonstrate its use for the classification of digits. Manuscripts with an accuracy comparable to that achieved by full-image reading, but with less delay and power consumption.

Find open access “Sparse pixel image sensor” spec sheet here. Published in April 2022.

Mennel, L., Polyushkin, DK, Kwak, D. et al. Sparse pixel image sensor. Sci Rep 12, 5650 (2022).

Visit the Semiconductor Engineering Technical Article Library here and discover many more academic articles about the chip industry.

Michael C. Garrison