# Compression-Efficient Feature Extraction Method for a CMOS Image Sensor

## TL;DR

This paper introduces a method to extract and compress image data efficiently for IoT devices, significantly reducing data size while maintaining high recognition accuracy.

## Contribution

A novel compression-efficient feature extraction method for CMOS image sensors that reduces data size by over 99% without significant loss in accuracy.

## Key findings

- The method achieves 60.7% APL50 accuracy on COCO with 99.2% less data than 8-bit RGB images.
- It also reaches 89.4% classification accuracy on VWW with 99.0% data reduction.
- The approach outperforms the conventional trade-off between data size and recognition accuracy.

## Abstract

To address the power constraints of the emerging Internet of Things (IoT) era, we propose a compression-efficient feature extraction method for a CMOS image sensor that can extract binary feature data. This sensor outputs six-channel binary feature data, comprising three channels of binarized luminance signals and three channels of horizontal edge signals, compressed via a run length encoding (RLE) method. This approach significantly reduces data transmission volume while maintaining image recognition accuracy. The simulation results obtained using a YOLOv7-based model designed for edge GPUs demonstrate that our approach achieves a large object recognition accuracy (APL50) of 60.7% on the COCO dataset while reducing the data size by 99.2% relative to conventional 8-bit RGB color images. Furthermore, the image classification results using MobileNetV3 tailored for mobile devices on the Visual Wake Words (VWW) dataset show that our approach reduces data size by 99.0% relative to conventional 8-bit RGB color images and achieves an image classification accuracy of 89.4%. These results are superior to the conventional trade-off between recognition accuracy and data size, thereby enabling the realization of low-power image recognition systems.