In [1]:

    

IMAGE_DIMS = (32,32,3)
INPUT_DIR = '../../input/kaggle-sea-lion/'
INPUT_DATASET_NAME = '02/lion-patches-0px'


    

In [6]:

    

%matplotlib inline
import cv2
import numpy as np
import pandas as pd
import h5py
import matplotlib.pyplot as plt
import sklearn
import os
import glob
import selectivesearch

from modules.logging import logger
import modules.utils as utils
from modules.utils import Timer
import modules.logging
import modules.cnn as cnn
import modules.lions as lions


    

In [ ]:

    

input_dataset_path = INPUT_DIR + utils.dataset_name(INPUT_DATASET_NAME, IMAGE_DIMS)
samples = None
with h5py.File(input_dataset_path, 'r') as input_h5:
    samples = input_h5['X'][20:40]
    utils.show_images(samples, cols=10, is_bgr=True, size=2)


    

In [ ]:

    

img_file = INPUT_DIR + 'Train/0.jpg'
im0 = cv2.imread(img_file)
im = im0[3000:4000,3000:4000,:]
utils.show_image(im, is_bgr=True)


    

In [ ]:

    

h5file = h5py.File(input_dataset_path, 'r')
xy_generator = utils.BatchGeneratorXYH5(h5file, start_ratio=0, end_ratio=1)
samples, _ = utils.dump_xy_to_array(xy_generator.flow(), xy_generator.size, x=True, y=False, dtype='uint8', random_skip=0.5)
print(np.shape(samples))
samples_hsv = [cv2.cvtColor(s,cv2.COLOR_BGR2HSV) for s in samples]


    

In [ ]:

    

im_hsv = cv2.cvtColor(im,cv2.COLOR_BGR2HSV)

# calculating object histogram
sample_hist = cv2.calcHist(samples_hsv,[0,1], None, [255,255], [0, IMAGE_DIMS[0], 0, IMAGE_DIMS[1]])
# sample_hist = cv2.normalize(sample_hist,0,255,cv2.NORM_MINMAX)

plt.hist(sample_hist)

#perform back projection of histogram
s = np.shape(im_hsv)
im_back = cv2.calcBackProject([im_hsv],[0,1],sample_hist,[0,s[0],0,s[1]],1)


    

In [ ]:

    

# Now convolute with circular disc
disc = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(5,5))
im_back_filter = cv2.filter2D(im_back,-1,disc)

utils.show_image(im_back)

# threshold and binary AND
_,mask = cv2.threshold(im_back_filter,50,255,cv2.THRESH_BINARY_INV)

mask = cv2.merge((mask,mask,mask))
utils.show_image(mask, size=6, is_bgr=True)

kernel = np.ones((3,3),np.uint8)
mask = cv2.erode(mask,kernel,iterations = 5)
mask = cv2.dilate(mask,kernel,iterations = 12)
utils.show_image(mask, size=6, is_bgr=True)

im_masked = cv2.bitwise_and(im_hsv,mask)
im_masked = cv2.cvtColor(im_masked,cv2.COLOR_HSV2BGR)

utils.show_image(im, size=6, is_bgr=True)
utils.show_image(im_masked, size=14, is_bgr=True)


    

In [ ]:

    

# params = cv2.SimpleBlobDetector_Params()
# params.minThreshold = 150;
# params.maxThreshold = 256;
# params.filterByArea = True
# params.minArea = 10
# params.filterByCircularity = False
# params.minCircularity = 0.1
# params.filterByConvexity = True
# params.minConvexity = 0.5
# params.filterByInertia = False
# params.minInertiaRatio = 0.5
# detector = cv2.SimpleBlobDetector_create(params)
# keypoints = detector.detect(im_masked)

# im_keypoints = cv2.drawKeypoints(im, keypoints, np.array([]), (0,0,255), cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# utils.show_image(im_keypoints, size=12, is_bgr=True)


    

In [ ]:

    

import skimage
regions = skimage.segmentation.felzenszwalb(im, scale=100, sigma=0.5, min_size=10)
utils.show_image(regions, size=6)


    

In [ ]:

    

print('selectivesearch')
img_lbl, regions = selectivesearch.selective_search(im_masked, scale=200, sigma=0.8, min_size=200)

print('filtering')
regions2 = []
for a in regions:
    size = a['size']
    r = a['rect']
    regions2 = regions2 + [r]


    

In [ ]:

    

print('showing')
im2 = im.copy()
print(len(regions2))
count = 0
for r in regions2:
    x, y, w, h = r
    if(w==0 or h==0):
        continue
    if(20<w<120 and 20<h<120 and 0.7<(w/h)<1.3):
        cv2.rectangle(im2, (x, y), (x+w, y+h), (0,255,0), 2)
        count += 1
print(count)
utils.show_image(im2, is_bgr=True, size=14)


    

In [ ]: