A saliency weighted no-reference perceptual blur metric for the automotive environment

This paper proposes a new approach for predicting the Quality of Experience (QoE) of fish-eye to rectilinear transformed images used in automotive vision applications. For this purpose a dataset of automotive images was created. Subjective image quality evaluations of the dataset were carried out, in terms of visual perception and driving assistance usefulness. For objective artifact description we have utilized some fundamental descriptors from the Fourier transform which are known to correlate well with perceptual blur. However, since the relevance of the detected artifacts is dependent on the image content saliency (visual perception focus), we optimize these measures for our application by locally weighting them according to visual saliency maps. The results show that radial to rectilinear conversion, which eliminates perspective distortion and maintains a similar field of view to that of the fisheye lens can be achieved with only minor loss in perceptual quality. Furthermore; it is shown that our algorithm, although relatively simple and computationally inexpensive, can accurately predict perceptual image quality in this environment, particularly for daytime driving conditions.