Region of Interest (ROI)
A photo contains far more visual information than a system usually needs at one moment. A Region of Interest (ROI) marks the particular part of an image or video frame worth examining—such as a face, a vehicle, a barcode, or a suspicious spot in an X-ray.
What an ROI is
An ROI is commonly represented as a rectangle using image coordinates: its starting position (x, y), plus its width and height. Cropping an ROI extracts only those pixels for later processing. In other tasks, an ROI can be a more precise shape: a polygon around a road lane or a pixel-level mask around an organ. The key idea is selection: tell the vision system where to focus rather than treating every pixel as equally relevant.
How it is used
ROIs can come from a person, a fixed camera setup, or another model:
- In a factory, engineers define an ROI around the product area and ignore the conveyor belt and background.
- In face recognition, a face detector first finds a bounding box; that box becomes the ROI sent to a recognition model.
- In autonomous driving, detected vehicle and pedestrian boxes are ROIs used for closer classification or tracking across video frames.
- In medical imaging, a clinician or segmentation model can isolate a lesion ROI for measurement and diagnosis support.
roi = image[y:y+h, x:x+w]. This is simple, but coordinates must stay within the image boundaries.
Why focus matters
Restricting work to an ROI reduces computation, removes distracting pixels, and can improve reliability when the relevant area is known. It also supports multi-stage detectors such as Faster R-CNN, which generates candidate regions and examines them in detail. A poorly chosen ROI can do the opposite: crop out part of the object, include misleading background, or cause a downstream model to miss the target entirely. An ROI is therefore not merely a crop—it is a decision about what visual evidence the system is allowed to use.
A Region of Interest (ROI) is a selected portion of an image or video frame chosen for focused analysis, processing, or annotation. It can be defined by a bounding box, mask, or coordinate range around relevant content, such as a face, organ, or vehicle. ROIs reduce computation and help vision systems concentrate on the pixels most relevant to detection, tracking, measurement, and recognition tasks.
Imagine looking at a busy photo and drawing a circle around the one thing you care about: a face, a car, a tumour on a scan, or a product on a shelf. That marked area is a Region of Interest (ROI).
In computer vision, an ROI is the part of an image or video that deserves special attention. Instead of treating every pixel as equally important, a system can focus on the relevant area. For example, a traffic camera may use an ROI around the road rather than the sky, or a medical tool may focus on a suspicious spot in an X-ray.
ROIs help AI pay attention to what matters, making visual tasks clearer and often more efficient.