Advanced cloud cover assessment

One or more non-transitory machine-readable mediums having stored thereon a sequence of instructions, which, when executed by a machine, cause the machine to automatically identify cloudy pixels in a target image, the target image comprising pixels associated with geographic coordinates, the geographic coordinates being associated with at least one feature, by performing the actions of:

• retrieving a cloud-free image from a cloud-free database, the cloud-free image corresponding with the target image and containing coincident geographic coordinates associated with a coincident feature;
• matching the at least one feature of the target image with the coincident feature of the cloud-free image;
• based on the matching, determining a cloud presence probability for each pixel in the target image;
• producing a probability raster image based on the cloud presence probability for each pixel in the target image;
• using the probability raster image, determining a cloud fraction, the cloud fraction being a ratio of high probability cloud pixels and total number of pixels;
• computing a confidence measure;
• comparing the cloud fraction against a predetermined cloud fraction threshold to obtain a cloud fraction result;
• comparing the confidence measure with a predetermined confidence measure threshold to obtain a confidence measure result;
• and
• using the cloud fraction result and the confidence measure result, determining whether the target image contains cloudy pixels.

A method for automatically determining whether a target image is substantially free of cloudy pixels, comprising:

• determining geographic coordinates of a bounding box of the target image;
• retrieving from a cloud-free data base a corresponding cloud-free image for the geographic coordinates of the target image;
• matching a feature in the bounding box of the target image with a corresponding feature in the bounding box of the cloud-free image by calculating a displacement magnitude between a plurality of pixels of the feature in the target image and a plurality of corresponding pixels of the corresponding feature in the cloud-free image, the displacement magnitude indicating a probability that the plurality of pixels are cloudy pixels;
• identifying a tie point associated with the plurality of pixels of the feature and the plurality of corresponding pixels of the corresponding feature, the tie point having a low cloudy pixel probability;
• establishing a tie point threshold;
• if a plurality of tie points meet or exceed the tie point threshold:
• comparing a first intensity of the target image to a second intensity of the cloud-free image;
• deriving a difference statistic from the intensity difference between the first intensity and the second intensity;
• deriving a probability statistic from the difference statistic, the probability statistic indicating a likelihood that the plurality of pixels are cloudy pixels;
• or
• if the plurality of tie points are below the tie point threshold, deriving an intensity probability from the first intensity of the target image;
• using the probability statistic or the intensity probability to compute a confidence measure;
• establishing a confidence threshold;
• comparing the confidence measure with the confidence threshold;
• classifying the pixel as a non-cloudy pixel if the confidence measure is below the confidence threshold or classifying the pixel as the cloudy pixel if the confidence measure meets or exceeds the confidence threshold;
• identifying a total number of cloudy pixels in the target image;
• computing a cloud fraction by dividing the total number of cloudy pixels in the target image by a total number of pixels in the target image;
• computing an overall confidence measure by summing up the confidence measures for each cloudy pixel in the total number of cloudy pixels and dividing by the total number of cloudy pixels in the target image;
• establishing a cloud fraction threshold and an overall confidence threshold;
• comparing the cloud fraction to the cloud fraction threshold and the overall confidence measure to the overall confidence threshold;
• if the cloud fraction falls below the threshold and the overall confidence measure meets or exceeds the overall confidence threshold, designating the target image as substantially cloud-free for inclusion into the cloud-free image database, the target image being a designated target image.

A method for identifying a digital image for a cloud-free database, comprising:

• providing a digital target image of a geographic location and a corresponding cloud-free digital image of the geographic location, the corresponding cloud-free image being retrieved from a cloud-free database;
• establishing a confidence threshold for the target image;
• for each pixel in a bounding box of the target image, automatically computing a confidence measure based on data from each corresponding pixel in the cloud-free image;
• automatically classifying the pixel as a cloudy pixel if the confidence measure meets or exceeds the confidence threshold;
• for each pixel in the bounding box of the target image, automatically applying a probability assessment to assign a cloud hypothesis value and a non-cloud hypothesis value to each pixel;
• automatically creating a raster image of the bounding box of the target image using the cloud hypothesis value and the non-cloud hypothesis value;
• establishing a cloudy pixel mask threshold;
• automatically comparing a value of each pixel in the bounding box of the target image to the cloudy pixel mask threshold;
• automatically setting to zero pixel values below the threshold and setting to 255 pixel values equal to or above the threshold, the result being a cloudy pixel mask;
• for the cloudy pixel mask, automatically determining a cloud fraction by dividing a total number of pixels set to 255 by a total number of pixels in the bounding box of the target image;
• automatically determining an overall confidence measure by summing up values for each pixel in the raster image for which a corresponding cloudy pixel mask value is 255 and dividing by a total number of cloudy pixels in the target image;
• and
• automatically designating the target image for inclusion in the cloud-free database if the cloud fraction is low and the overall confidence measure is high, the target image being a designated target image.

One or more non-transitory machine-readable mediums having stored thereon a sequence of instructions, which, when executed by a machine, cause the machine to automatically determine whether a target image is substantially free from cloudy pixels, the target image comprising a bounding box associated with geographic coordinates, by performing the actions of:

• retrieving from a cloud-free data base a corresponding cloud-free image for the geographic coordinates of the target image;
• matching a feature in the bounding box of the target image with a corresponding feature in the bounding box of the cloud-free image by calculating a displacement magnitude between a plurality of pixels of the feature in the target image and a plurality of corresponding pixels of the corresponding feature in the cloud-free image, the displacement magnitude indicating a probability that the plurality of pixels are cloudy pixels;
• identifying a tie point associated with the plurality of pixels of the feature and the plurality of corresponding pixels of the corresponding feature, the tie point having a low cloudy pixel probability;
• if a plurality of tie points meet or exceed a predetermined tie point threshold:
• comparing a first intensity of the target image to a second intensity of the cloud-free image;
• deriving a difference statistic from the intensity difference between the first intensity and the second intensity;
• deriving a probability statistic from the difference statistic, the probability statistic indicating a likelihood that the plurality of pixels are cloudy pixels;
• if the plurality of tie points are below the tie point threshold, deriving an intensity probability from the first intensity of the target image;
• using the probability statistic or the intensity probability to compute a confidence measure;
• comparing the confidence measure with a predetermined confidence threshold;
• classifying the pixel as a non-cloudy pixel if the confidence measure is below the confidence threshold or classifying the pixel as the cloudy pixel if the confidence measure meets or exceeds the confidence threshold;
• identifying a total number of cloudy pixels in the target image;
• computing a cloud fraction by dividing the total number of cloudy pixels in the target image by a total number of pixels in the target image;
• computing an overall confidence measure by summing up the confidence measures for each cloudy pixel in the total number of cloudy pixels and dividing by the total number of cloudy pixels in the target image;
• comparing the cloud fraction to a predetermined cloud fraction threshold and the overall confidence measure to a predetermined overall confidence threshold;
• if the cloud fraction falls below the threshold and the overall confidence measure meets or exceeds the overall confidence threshold, designating the target image as substantially cloud-free for inclusion into the cloud-free image database.