Global and Local Maximas of the histogram

Question

I'm trying to segment the football field. I'm following a research paper that suggests

Convert to HSV
Grab the Hue Channel
Generate Histogram
Grab Global and local maxima (Global max can be max(hist))
If a local maxima is of 20% of the Global Maxima, Consider it

The issue is that I don't know how to get Local Maximas. I'm trying to capture a list of peaks in the histogram.

I have tried it on MatLab, it works quite well but I need to do it in python. I have tried libraries like peakutils but nothing is giving me desired results.

def field_area_mask(image):

    # Convert to HSV
    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)

    # Capture the Hue Channel
    hue = hsv[:, :, 0]

    # Generate Histogram
    hist = cv2.calcHist([hue],[0],None,[256],[0,256])

    # Capture range
    hist_range = hist[:121] # 0-120
    hist_range = hist_range.reshape(1, -1)[0]
    Hmax = max(hist_range)
    plt.plot(hist_range)

    HiMax = [] # Need to populate the local maxima's list

*[Finding local maxima/minima with Numpy in a 1D numpy array](https://stackoverflow.com/questions/4624970/finding-local-maxima-minima-with-numpy-in-a-1d-numpy-array)* might help you. You can also have a look at [`argrelextrema`](https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.argrelextrema.html) function from sklearn module. — Alexandre B., Jul 26 '19 at 12:56
See http://stackoverflow.com/questions/3790445/1d-multiple-peak-detection — fmw42, Jul 27 '19 at 01:26

score 1 · Answer 1 · answered Jul 27 '19 at 00:49

1

See numpy.diff

From 1st derivative you can find stationary points. From 2nd derivative you can tell if these points are local max or min.

Example: given your histogram

[1, 2, 1, 3, 7, 10, 7, 2] // 0-based index

1st derivative is

[1, -1, 2, 4, 3, -3, -5] // 1-based index

sign changes at 2nd, 3rd and 6th element.

2nd derivative is

[-2, 3, 2, -1, -6, -2] // 2-based index

sign for 2nd element is (-), local max

sign for 3rd element is (+), local min

sign for 6th element is (-), local max

global max is 10, so your threshold is %20 * 10 = 2. 2nd element shall be accepted as a local max.

answered Jul 27 '19 at 00:49

baci

2,528
15
28

Not always so simple. You have to suppress all the points on the sides of the main peak that are higher than any other local max, etc. See http://stackoverflow.com/questions/3790445/1d-multiple-peak-detection – fmw42 Jul 27 '19 at 01:23
See also https://stackoverflow.com/questions/5550290/find-local-maxima-in-grayscale-image-using-opencv – fmw42 Jul 28 '19 at 00:35

score 1 · Answer 2 · answered Jul 27 '19 at 10:27

1

I was able to find the peaks using the scipy.signal's method argrelextrema. I used it with np.greater.

peaks = argrelextrema(hist_range, np.greater)

answered Jul 27 '19 at 10:27

Ibtihaj Tahir

636
5
17

Global and Local Maximas of the histogram

2 Answers2