`ValueError` when computing statistics on ROI extending beyond image boundaries

[PierreRaybaut]

GitHub Issue: ROI Statistics with Out-of-Bounds ROI

Title

Fix ValueError when computing statistics on ROI extending beyond image boundaries

Labels

• bug
• image processing
• ROI

Description

Problem

When computing statistics (or any feature) on an image with a ROI that extends beyond the canvas boundaries, a ValueError is raised:

ValueError: zero-size array to reduction operation minimum which has no identity

Steps to Reproduce

1. Create an image (e.g., 100x100 pixels)
2. Define a ROI that extends beyond the image canvas (e.g., rectangle at x=80, y=80 with width=30, height=30, extending to x=110, y=110)
3. Compute statistics on the image with this ROI
4. Error is raised during statistics computation

Root Cause

The ImageObj.get_data() method did not clip ROI bounding box coordinates to image boundaries. When a ROI extended beyond the canvas, the array slice could result in an empty array (zero-size), causing numpy masked array functions to fail.

Solution

Modified ImageObj.get_data() method in sigima/objects/image/object.py to:

• Clip bounding box coordinates to image boundaries using max(0, ...) and min(..., shape)
• Handle completely out-of-bounds ROIs by returning a fully masked 1x1 array (containing NaN)
• Properly handle partial overlap cases

Test Cases

Three scenarios tested:

1. ROI partially extending beyond canvas (right/bottom) - ✅ Works
2. ROI completely outside canvas - ✅ Returns NaN values
3. ROI with negative coordinates (extends beyond top/left) - ✅ Works

Files Changed

• sigima/objects/image/object.py - Fixed ImageObj.get_data() method
• CHANGELOG.md - Documented the fix

Backward Compatibility

This fix maintains backward compatibility. Existing code will continue to work, and ROIs that were previously within bounds behave identically.