On the influence of using quantized images on machine learning algorithms
Takayuki Osakabe, Yuma Kinoshita, Hitoshi Kiya · 2020
Recently, applying quantized images to machine learning algorithms has been expected to enhance robustness against adversarial examples. In this paper, three quantization methods: linear quantization, lloyd-max quantization and error diffusion are considered for producing quantized images, respectively, and the influence of using the quantized images is discussed in an image classification experiment under the use of typical machine learning algorithms including deep learning ones.