simple program for reading an image, finding SURF interest points, clustering those, determining the centers of the clusters, and drawing circles around these points.

analyze_image.py

#!/usr/bin/env python3

'''
Usage:
    
    analyze_pic [image] [mode]

    modes:
        0 : grayscale
        1 : color

'''

# imports
import cv2
import sys
import matplotlib.pyplot as plt 
import seaborn as sns; sns.set()
from sklearn.cluster import KMeans

def analysis(img):
    '''all the logic and processing for the program'''

    # initial analysis
    surf = cv2.xfeatures2d.SURF_create(400)
    kp, des = surf.detectAndCompute(img, None)

    # determines the number of clusters to be searched for
    k_clusters = int(len(kp)/100)

    if k_clusters < 10:
        k_clusters = 10
    elif k_clusters > 40:
        k_clusters = 40
    
    # writes original image
    cv2.imwrite(name + "_" + mode + "_1.jpg", img)

    # extracts key points and stores them for the k-means
    # algorithm to read cleanly    
    kp_dict = {i:kp[i].pt for i in range(len(kp))}
    kp_x = [i[0] for i in kp_dict.values()]
    kp_y = [i[1] for i in kp_dict.values()]

    key_points = [[kp_x[i], kp_y[i]] for i in range(len(kp_x))]

    # kmeans algorithm for determining cluster center points
    kmeans = KMeans(n_clusters = k_clusters)
    kmeans.fit(key_points)

    centers = kmeans.cluster_centers_

    # draws circles on the image to visualize the keypoints
    for i in range(len(centers[:, 0])):
        cv2.circle(img, (int(centers[:, 0][i]), int(centers[:, 1][i])), 50, 1000)
        print(centers[:, 0][i])

    # writes edited image with keypoints
    cv2.imwrite(name + "_" + mode + "_2.jpg", img)

# reads user input
name = sys.argv[1].split(".")[0]
mode = sys.argv[2]

img = cv2.imread(sys.argv[1], int(mode))

# runs main logic on the input image
analysis(img)