Fig 1.
The radiance reaching the smartphone camera is the summation of the transmitted light from the object and airlight from the sun after scattering by air, water and PM in atmosphere.
Fig 2.
The histogram of PM2.5 in different cities.
(a) Beijing; (b) Shanghai; (c) Phoenix.
Fig 3.
PM estimation via outdoor image analysis.
Fig 4.
Sample photos in our haze detection database.
a) Photos captured at Beijing, Shanghai and Phoenix respectively. b) Boundary lines (blue lines in b) between distant buildings and sky. c) Selected ROIs (red boxes). Reprinted under a CC BY license, with permission from [Yi Zou], original copyright [2014].
Fig 5.
The transmission decreases as the distance or PM2.5 index increases.
(a) Schematic illustration of transmission variation with distance. (b) Four ROIs (r1~r4) located at increasing distances. (c) The estimated transmission map. (d) Semi-logarithmic plots of transmission curves vs. distance under different haze conditions.
Fig 6.
Image features variation with PM index.
(a~d): Hazy images showing that the contrast of the building region decreases with the PM index, where the lower panel shows the zooming-in images of the regions marked by the red boxes. (e) The normalized features vs. PM2.5 index plot, including ROI RMS contrast (blue), image entropy (black), and image RMS contrast (red).
Fig 7.
Sky gradient and blue component provide weather information, such as cloud formation.
(a) Sunny day; (b) Partly cloudy/sunny day; (c) Hazy day; (d) Cloudy day.
Fig 8.
Sky color dependence on solar zenith angle.
(a) Definition of solar zenith angle. (b) Sample images show that the sky near horizon is red during sunrise and sunset on the same day compared with noon time.
Fig 9.
Real PM2.5 index vs. predicted PM2.5 index plot.
(a) Beijing; (b) Shanghai; (c) Phoenix.
Table 1.
Assessment of the support vector regression.
Table 2.
Regression results for different weather conditions.
Table 3.
Regression results with and without humidity as a feature.
Table 4.
The features and their correlations with PM2.5 index in our dataset.
Fig 10.
PCA-SVR results for Beijing and Shanghai’s dataset.
(a) RMSE. (b) R-squared.
Table 5.
The performance comparison between all the features and the optimized feature subsets for Beijing and Shanghai’s dataset.