Detection of sky in digital color images
View Patent ↗A method of detecting sky in a digital color image having pixels is disclosed. The method includes identifying pixels from the digital color image representing an initial sky region; developing a model based on the identified sky pixels, wherein such model is a mathematical function that has inputs of pixel position and outputs of color; and using the model to operate on the digital color image to classify additional pixels not included in the initial sky region as sky.
1. A method of detecting sky in a digital color image having pixels, the method comprising:
a) identifying pixels having a range of colors from the digital color image and forming an initial sky region using the identified sky pixels;
b) developing a model based on the identified sky pixels, wherein such model is a mathematical function that has inputs of pixel position and outputs of color;
c) using the model to operate on the digital color image to classify additional pixels not included in the formed initial sky region as sky including;
i) determining a difference between the color of a pixel and the model estimated color for the pixel's position;
ii) using the determined difference at a plurality of pixel positions to calculate a residual error from the differences; and
iii) using the difference to assign a probability that the pixel represents sky further, by computing a ratio of the difference with the residual error and using the ratio to assign the probability; and
d) using the probability to determine if the pixel is in a sky region.
2. The method of claim 1 wherein the model is a two-dimensional polynomial of the pixel position in the digital color image.
3. The method of claim 2 wherein the polynomial is a second-order polynomial.
4. The method of claim 1 wherein the step of identifying an initial sky region further comprises:
i) extracting connected components of said potential sky pixels;
ii) eliminating ones of said connected components that have a texture above a predetermined texture threshold; and
iii) computing desaturation gradients of said connected components; and comparing said desaturation gradients of said connected components with a predetermined desaturation gradient for sky to identify true sky regions in said digital color image.
5. The method of claim 1 wherein in computing the probability of the pixels as sky further comprises:
computing a color feature, related to the “blueness” of the pixels in the region; and
using the color feature to determine if the pixels are not sky.