Cloud Shadow Removal Software for Aerial imaging and Photogrammetry

Cloud shadows remain one of the persistent challenges in aerial imaging and photogrammetry. As clouds intersect with the sun’s rays, their shadows are cast onto the landscape below, an unavoidable…

Cloud shadows remain one of the persistent challenges in aerial imaging and photogrammetry. As clouds intersect with the sun’s rays, their shadows are cast onto the landscape below, an unavoidable effect that frequently appears in captured imagery. Before such data can be used for mapping and 3D visualization, these shadows must be removed.

Traditionally, cloud shadow removal in image processing relies on creating masks and applying filters to specific regions of an image, such as shadowed areas. In some cases, objects that should remain unaffected by filtering are isolated, often across different layers. However, this workflow is largely manual, making it slow and prone to inaccuracies. The task is further complicated by the fact that shadows often do not have clearly defined boundaries, which makes precise masking and seamless correction difficult.

The Aerial Clarity algorithm addresses this issue using classical image processing principles, without relying on machine learning or artificial intelligence. It is specifically designed for soft shadows, where explicit boundaries may be absent. This situation is common in aerial and photogrammetry imagery, where shadow removal is typically required before further processing steps.

According to its developers, parallel computation on an NVIDIA GeForce RTX 4090 enables performance levels of approximately 5–7 GPix/s or higher for high-resolution 16-bit RGB images. Such performance, they note, is not achievable with AI-based systems.

It is emphasized that the algorithm is applicable only to soft cloud shadows and is not intended to suppress deep or hard shadows.

Key Issues

• Automatic cloud shadow removal for aerial images (soft shadows only)
• No limitation on image resolution, 245.7 MPix (IMX811 with resolution 19,200 × 12,800) is ok
• High performance due to parallel algorithm implementation on CUDA

Classic vs AI-based Shadow Removal

Unlike machine learning approaches which require training data and often struggle with generalized shadow types, this classic computer vision approach relies solely on pixel-level image characteristics, statistics and deterministic filtering. This results in:

• Pros:
  • Fast GPU execution, even for very high image resolutions
  • No need for data gathering, annotating and training, no dependency on large labeled datasets
  • Robust for soft and diffuse shadows, which are typical in aerial imagery and photogrammetry
• Cons:
  • Not applicable to deep shadows
  • Shadow removal may result in the presence of minor traces on the image
  • Less adaptive to complex, varying shadow types compared to AI methods

AI-based shadow removal, while potentially more flexible, often requires extensive training and may struggle to generalize to unseen scenes or shadow types. For large-scale aerial and photogrammetry applications where speed and robustness are critical, classic methods provide an efficient solution.

More information, https://www.fastcompression.com/