Find Reference

This function block locates a single instance of a reference image inside a target image, estimates its position and orientation, and returns visual and geometric results that help you inspect or further process the detected object.

📥 Inputs

Object Image The image in which the reference should be located (target image).

Reference Image The small reference image used to search inside the object image.

Reference Image ROI Optional shape input to restrict matching to a region inside the reference image (useful to ignore irrelevant parts).

📤 Outputs

Result Image The object image annotated with the detected reference bounding polygon.

Detected Object Image A perspective-corrected crop of the detected object, transformed to match reference orientation/size.

Keypoints Image A visualization of feature/keypoint detections on the object image to help verify matches.

Bounding Box The enclosed rectangle that bounds the detected reference in the object image.

Corner Coordinates The four polygon corner coordinates of the detected reference area.

Center Position The center point (x, y) of the detected reference.

🕹️ Controls

This block does not expose additional widget controls beyond its inputs. Use the Reference Image ROI input to limit the reference area when needed.

🎨 Features

• Visual detection result with polygon overlay for quick verification.

• Returns both a perspective-corrected crop and geometric outputs (corners, bounding box, center) for downstream processing.

• Accepts an optional ROI on the reference image to focus matching on a sub-region.

• Produces a keypoints visualization to help debug matching quality.

📝 Usage Instructions

1. Provide the source image to Object Image and the template/reference to Reference Image.

2. Optionally connect a shape to Reference Image ROI to target a specific area inside the reference.

3. Run the scenario — the block will attempt to find the reference and output annotated visuals and geometry.

4. Use the outputs to draw overlays, crop, measure or feed into decision logic.

📊 How it runs

When executed, the block compares distinctive visual features between the reference and object images, estimates the position and orientation of the best match, and returns a visual annotation plus transformed and numeric geometry outputs that represent the found object.

💡 Tips and Tricks

• To focus detection on a specific area of the input image, crop beforehand with Image ROI Select or Image ROI and feed the cropped image into Object Image.

• If scale or orientation differs significantly between reference and object, try resizing the inputs with Image Resize before matching.

• Improve robustness under varying lighting by preprocessing with Adjust Colors, Contrast Optimization, or Blur to reduce noise.

• Use Show Image to preview Result Image or Detected Object Image interactively while tuning inputs.

• Combine the block output with Draw Detections or Draw Result On Image to overlay textual or rectangular annotations on live images.

• If you need a perspective matrix for other measurements, consider following up with Perspective Transform using the detected corner coordinates.

• When the match returns many false positives, try limiting the reference features using Reference Image ROI or apply Image Threshold on inputs to simplify textures.

(hint) For region-based workflows, use Get ROI after detection to extract and forward the detected area to other processing blocks.

🛠️ Troubleshooting

• No detection found: ensure the reference appears in the object image with sufficient visible texture; try increasing the reference size in the object or use Image Resize.

• Poor or unstable matches: improve lighting and contrast with Adjust Colors or Contrast Optimization; reduce noise with Blur.

• Detection located but warped output looks wrong: validate the corner coordinates using Show Image on the Keypoints Image and consider restricting the Reference Image ROI.

• Slow processing on large images: downscale with Image Resize before feeding into this block, then map results back to original coordinates if needed.