opencv and python: how to use cv2.surf() with mask

问题

I am a newbie at opencv and python and am trying to collect keypoints and descriptors of faces within an image.

I am using HAAR cascade classifier with frontal face template to look for faces in an image. The HAAR cascade gives me a list of coordinates marking the faces in the image. I want to generate a "mask" at those coordinates so that I can use cv2.surf() to extract keypoints and descriptors within the masked region.

I don't know how to create that mask.

Try this photo as an example to work on.

Here is the code thus far:

import cv2
import numpy as np

# Load image and convert to grayscale
img = cv2.imread('testPhoto.jpg')
imgg = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Look for faces in the image
cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt.xml')
faceRegions = cascade.detectMultiScale(imgg)

After this I'd like to do the SURF extraction with a mask using faceRegions. Suppose faceRegions looks like this:

array([[488, 163,  91,  91],
       [357, 184,  93,  93],
       [154,  78, 107, 107]], dtype=int32)

There were three faces found in imgg so I want to create three separate masks at their specific locations with their particular width and height. And then have cv2.surf() look only in the masked regions. How can I do that?

回答1:

The faceRegions you got denote the x,y,width,height of the faces. So you can simply set a ROI (Region of Interest) with these coordinates and send that rectangle as an image to your SURF function.

For eg: