Why doesn't doesn't SIFT work for 8 bit images (JPEG) in OpenCV with Python?

Question

I used SIFT for all my other 24 bit JPEG images without any problems, however, the 8-bit one always give me the following error.

image is empty or has incorrect depth (!=CV_8U) in function cv::SIFT::operator ()

Does anyone know how to deal with it?

Here is my code:

import cv2 
import numpy as np 
import os 
import glob
import scipy.cluster
os.chdir('\mydirectory')
images = []

for infile in glob.glob('./*.jpg'):
  pic = cv2.imread(infile,0)
  images.append(pic)

my_set = images
descriptors = np.array([])
feaL=np.array([])

for pic in my_set:
  kp, des = cv2.SIFT().detectAndCompute(pic, None)
  feaL=np.append(feaL,des.shape[0])
  descriptors = np.append(descriptors, des)

Then the error "image is empty or has incorrect depth (!=CV_8U) in function cv::SIFT::operator ()" pops up.

Answer 1

EDIT: After typing this I just saw the grayscale flag on imread. Try printing the images as they are read in, it sounds like imread may be silently failing and leaving you with empty Mats.

cv2.SIFT.detectAndCompute never takes anything other than 8-bit grayscale, so I'm not sure that you actually did use SIFT on a 24-bit image without problems.

cv2.SIFT.detectAndCompute

Python: cv2.SIFT.detectAndCompute(image, mask[, descriptors[, useProvidedKeypoints]]) → keypoints, descriptors

So to change to 8 bit grayscale immediately prior to detection and extraction:

for pic in my_set:
    pic = cv2.cvtColor(pic, cv2.COLOR_BGR2GRAY)
    kp, des = cv2.SIFT().detectAndCompute(pic, None)

Of course that is a dumb place to put it, but it's up to you to figure out if you need to keep the BGR originals or not, etc.