Check Image Data Quality


UW-MOS task checking the quality and suitability of a set of images for 2D mosaicing and/or 3D reconstruction.

The first step in both 2D mosaicing and 3D reconstruction tasks is to detect features on the images. These features are the base of the subsequent steps in both pipelines and, consequently, the input images must contain a relevant number of such features in order to obtain good results.

This task computes the number of AKAZE [Alcantarilla2011] features for each image of the input dataset. A minimum of 500 features is required to consider an image as suitable for mosaicing/reconstruction.


The current version of the service only requires the user to upload a set of images.


Upon task completion, clicking on the corresponding See the results button on the task list will show a short report on the quality of the data.

The report indicates the number of images for which it was able to extract more than 500 features from the total of images uploaded. Additionally, it shows a list with the number of features extracted for each individual image, marking in red those images with less than 500 features.

Note that finding images in a dataset that fail this check does not mean they will not be suitable for mosaicing or reconstruction. This check is supposed to be just a mean for the user to detect those images that MAY suppose a problem for reconstruction due to lack of features, as this fact is usually related to blurred images, images with no distinctive texture content, etc. However, if they are relevant, the features found may be enough for building the mosaic or the 3D.


You can use the datasets for mosaicing and 3D reconstruction demos as input to this task.

Just upload the files and wait for the task to finish, a report similar to the one shown below should appear after clicking the See the results button:

Check quality demo results

Example of a report of the demo for the check quality task.



Pablo F Alcantarilla, Jesús Nuevo, and Adrien Bartoli. Fast explicit diffusion for accelerated features in nonlinear scale spaces. Trans. Pattern Anal. Machine Intell, 34(7):1281–1298, 2011.