The example in the mentioned link uses "extractFeatures" function to detect the objects in the target image and "showMatchedFeatures" function to match the features with detected object. In the same example, "estimateGeometricTransform" calculates the transformation relating the matched points, while eliminating outliers. This transformation allows us to localize the object in the scene.
If the number of matched features after above transformation is relatively high, you can be almost sure that the object is present in the target image and vice versa. The pixels within the object border specified by 'newBoxPolygon' variable can be read from the target image for further processing.