Qiita Teams that are logged in
You are not logged in to any team

Log in to Qiita Team
Community
OrganizationAdvent CalendarQiitadon (β)
Service
Qiita JobsQiita ZineQiita Blog
17
Help us understand the problem. What is going on with this article?
@jellied_unagi

Scikit-ImageのHOGの出力について

More than 5 years have passed since last update.

skimage.feature.hog (http://scikit-image.org/docs/dev/api/skimage.feature.html#hog) は入力画像からHOG特徴量を抽出してくれるのだが,出力結果は1d-arrayであり,何がなんだかわからないという問題がある.https://github.com/holtzhau/scikits.image/blob/master/skimage/feature/hog.py にあるコードを読むと,このarrayをどう解釈すれば良いかがわかる.

入力引数で重要なもの

  • orientations: ヒストグラムのビン数(default=9)
  • pixels_per_cell: セルのサイズ(default=(8,8))
  • cells_per_block: ブロックごとのセル数(default=(3, 3))

出力を見てみる

180行目

hog.py
return normalised_blocks.ravel()

ここravelせずに返して欲しかった.normalised_blocksの宣言を見てみると,

160行目

hog.py
n_blocksx = (n_cellsx - bx) + 1
n_blocksy = (n_cellsy - by) + 1
normalised_blocks = np.zeros((n_blocksx, n_blocksy, bx, by, orientations))

ここで使われている変数は100行目あたりで宣言されている

hog.py
sx, sy = image.shape
cx, cy = pixels_per_cell
bx, by = cells_per_block

n_cellsx = int(np.floor(sx // cx))  # number of cells in x
n_cellsy = int(np.floor(sy // cy))  # number of cells in y

というわけで,hogを回して得られる1-d array(retvalとする)は,

retval.reshape((n_blocksx, n_blocksy, bx, by, orientations))

としてあげると,元の形に戻すことができる.

活用例

たとえば,blockごとの最大勾配方向を可視化することができる.

visualize_argmaxhog.py
from skimage.feature import hog
from skimage.data import camera
import matplotlib.pyplot as plt

img = camera()
orientations = 9
pixels_per_cell = (8, 8)
cells_per_block = (3, 3)

h = hog(img, orientations=orientations, pixels_per_cell=pixels_per_cell, cells_per_block=cells_per_block)

sx, sy = img.shape[:2]
cx, cy = (8, 8)
bx, by = (3, 3)
n_cellsx = int(np.floor(sx // cx))  # number of cells in x
n_cellsy = int(np.floor(sy // cy))  # number of cells in y
n_blocksx = (n_cellsx - bx) + 1
n_blocksy = (n_cellsy - by) + 1

himg = h.reshape((n_blocksx * n_blocksy, bx, by, orientations))
vis = np.array([np.argmax(x.sum(axis=(0, 1))) for x in himg]).reshape((n_blocksx, n_blocksy))

plt.subplot(1, 2, 1)
plt.imshow(img)
plt.axis('off')
plt.subplot(1, 2, 2)
plt.imshow(vis, interpolation='nearest')
plt.axis('off')

kobito.1422522927.516947.png

17
Help us understand the problem. What is going on with this article?
Why not register and get more from Qiita?
  1. We will deliver articles that match you
    By following users and tags, you can catch up information on technical fields that you are interested in as a whole
  2. you can read useful information later efficiently
    By "stocking" the articles you like, you can search right away
jellied_unagi
Python and Matlab tips for computer vision

Comments

No comments
Sign up for free and join this conversation.
Sign Up
If you already have a Qiita account Login
17
Help us understand the problem. What is going on with this article?