Matching parts of an image from another image

I have an input image of an aeroplane and an input of 100 images with 10 of them being planes(similar to input).

I have generated bag of visual words from the input dataset with k = 500 and SIFT features.

If I give part of my input image (Say a cropped image containing one wing, or only the nose of the plane), then retrieving similar images from the dataset doesn't give any of the 10 images of the plane in top-20 or even in top-50 results. Though if I give the entire image as input, the retrieval performance is good.

Is there a way to match bag of visual words so that if only a part of an image in the dataset matches the entire input image, that may also be retrieved ? Are there any alternative methods to do so ?