Automatic Alignment for Image Comparisons

Dermi Atlas includes automatic image alignment, a computer vision-powered capability that spatially registers clinical photographs within split and overlay comparisons. When activated, corresponding anatomical features across images are detected and matched, and spatial transformations are applied so that anatomy lines up precisely between panels or layers. This enables more reliable visual assessment of changes between visits.

Alignment is powered by Atlas Vision, a local processing service that runs entirely on-premises as part of the Dermi Atlas Professional deployment. No patient images are transmitted to external servers during alignment.

How Automatic Alignment Works

The alignment process is triggered on demand by pressing the align button within a comparison. Once activated, the Atlas Vision service analyzes the images and performs the following steps:

1. Anatomical landmarks and visual features are detected in each image independently.
2. Corresponding features between images are matched, and a spatial transformation (translation, scale, and rotation) is computed for each image relative to the reference.
3. The transformation is applied to reposition each image so that matching anatomy is aligned across all panels or layers.

The entire process runs automatically. No manual landmark selection or positioning is required. The computed alignment is applied with a smooth animated transition, and the original image positions can be restored at any time by resetting the alignment.

Supported Comparison Modes

Automatic alignment is available in both comparison modes offered by Dermi Atlas.

Split Comparison

In split comparison mode, alignment adjusts images across all layout types (vertical, horizontal, and grid). Each image is repositioned and scaled so that the corresponding anatomical region occupies the same relative position in each panel. Synchronized viewing controls continue to function normally after alignment is applied.

Overlay Comparison

In overlay comparison mode, alignment transforms each layer so that anatomical features overlap precisely when layers are stacked. This makes opacity-based blending between images significantly more effective, as corresponding anatomy occupies the same pixel coordinates. Subtle differences in lesion size, shape, or pigmentation become more apparent when layers are properly registered.

Best Practices for Optimal Alignment

Alignment accuracy depends on the quality and consistency of the input images. The following practices are recommended for best results:

• Images should be captured from a consistent angle relative to the anatomical region of interest.
• A similar distance between the camera and the subject should be maintained across visits.
• Consistent lighting conditions reduce visual differences that are unrelated to clinical changes.
• The region of interest should be clearly visible and occupy a significant portion of the frame.
• Images with extreme differences in zoom level, angle, or cropping may produce less precise alignment.

Understanding Alignment Results

When alignment is applied, each image may be translated (shifted), scaled (resized), and rotated to achieve the best spatial match. These adjustments are applied non-destructively to the comparison view only; original image files are never modified.

Alignment state is automatically saved as part of the comparison session. If a comparison is closed and reopened later, the alignment configuration is preserved. The alignment can also be reset to restore the original, unaligned image positions.

Multi-Scale Alignment

Automatic alignment supports images captured at different zoom levels. For example, a wide-angle photograph from one visit and a close-up from another visit can be aligned and scaled so that the same anatomical region is displayed at a comparable size. This is particularly useful when imaging protocols vary between appointments or when different camera equipment is used.

Limitations and Expected Behavior

The following limitations should be considered when using automatic alignment:

• Alignment requires sufficient visual detail in the images. Very uniform or featureless regions (such as large areas of solid color) may not produce reliable matches.
• Extreme differences in perspective or viewing angle between images may result in reduced alignment accuracy. The system computes 2D similarity transforms (translation, scale, and rotation) and does not account for 3D perspective distortion.
• Alignment quality may be affected if significant portions of the anatomy are occluded or cropped differently between images.
• Very small images or heavily compressed images may contain insufficient detail for feature detection.
• In cases where alignment cannot be computed, images will remain in their original positions and no transformation will be applied.

Availability

Automatic alignment is available in Dermi Atlas Professional. All alignment processing is performed locally by the Atlas Vision service container, with no external network dependencies.