cv_sampling_by_hsv.py

# -*- coding: utf-8 -*-

import numpy as np
import cv2
import math

def rgb_to_hsv(r, g, b):
  hsv = cv2.cvtColor(np.array([[[b, g, r]]], dtype=np.uint8), cv2.COLOR_BGR2HSV)[0][0]
  return (hsv[0], hsv[1], hsv[2])

def main():
  # 画像を読み込む
  img = cv2.imread('images/sample_7.jpg')

  # 幅500pxを基準にリサイズ
  scale = 500 / img.shape[1]
  resized_img = cv2.resize(img, (round(img.shape[1] * scale), round(img.shape[0] * scale)))
  height, width, channels = resized_img.shape

  # 縦横の分割数とセルのサイズを決定
  div_y = 16
  div_x = 16
  cell_height, cell_width = round(height / div_y), round(width / div_x)

  # 結果用の画像を生成
  canvas = np.zeros([cell_height * div_y, cell_width * div_x, channels], 'uint8')

  for y in range(div_y):
    for x in range(div_x):

      pos_y = cell_height * y
      pos_x = cell_width * x
      
      # 平均化対象のセル
      target = resized_img[pos_y:pos_y + cell_height, pos_x:pos_x + cell_width]

      # 平均化
      averaged_color_per_row = np.average(target, axis=0)
      averaged_color = np.average(averaged_color_per_row, axis=0).astype('uint8')
      averaged_img = np.array([[averaged_color] * cell_width] * cell_height, np.uint8)

      # HSVに変換
      hsv = rgb_to_hsv(averaged_color[2], averaged_color[1], averaged_color[0])

      # 閾値以上のセルを描画する
      if hsv[0] in range(11, 100) and hsv[1] > 60 and hsv[2] > 80:
        canvas[pos_y:pos_y + cell_height, pos_x:pos_x + cell_width] = averaged_img
    
  # ウインドウに表示
  cv2.imshow('original', resized_img)
  cv2.imshow('processed', canvas)

  cv2.waitKey(0)
  cv2.destroyAllWindows()

if __name__ == '__main__':
  main()