Image enhancement and fast recognition of low-quality QR codes
The current research aim is to construct a fast system to enhance the quality of QR Code images to improve the decoding success rate, hence decreasing the time required to read the data contained in 2D bar codes. The system is an Convolutional Neural Network (CNN) with a structure of an auto-encoder, trained with a TensorFlow toolkit on self-generated images. The input is an image with a poor resolution and quality and the output is a clear, noise-less picture of the same QR Code. The output of the CNN can then be used to decode the QR Code’s data with any preferred decoding algorithm. The system has been deployed on the Raspberry Pi 3B+ and evaluated in different simulation obtaining an increase of about +100% success rate of correct decoding of low-quality QR Code images.
| System | 2.7 | 3.5 | 3.6 |
|---|---|---|---|
| Linux CPU | — | ||
| Linux GPU | — | ||
| Windows CPU / GPU | — | — | |
| Linux (ppc64le) CPU | — | — | |
| Linux (ppc64le) GPU | — | — |
- Image De-Blur and Enhancement
The enhancer developed is based on a Convolutional Neural Network or in short CNN, trained from zero and built on a structure usually referred as auto-encoder. CNNs are particular kinds of Neural Networks (NN) which simulate the convolution operation. The convolution I\*K between an input matrix I and another matrix, the kernel or filter, K is obtained by translate pixel-by-pixel the kernel
