Clustering in Python-Image Clustering
I want to group images using K Means or some other algorithm (suggestion required).
The problem is this: I want to copy images into 3 clusters (nature, sunset, water). I loaded all the images using os.listdir () and then converted all images to arrays (RGB) and then created a data frame that contains three columns - ID, Image_array, Label.
Now when I use KMans clustering by providing n_clusters = 3 it shows this error:
from sklearn.cluster import KMeans kmeans = KMeans (n_clusters = 3) .fit (img_array) ERROR = An array with dim 4 found. Expected expected evaluator <= 2.
Now I need your help on this clustering issue. The dataframe I created looks like this:
img_array = []
path = "C://Users/shivam/Desktop/freelancer/p22/data/green_nature/"
for f in os.listdir('.'):
if f.endswith('.jpg'):
img = Image.open(f)
data = np.asarray(img, dtype='uint8')
img_array.append(data)
df = pd.DataFrame({'image_arrays':img_array})
df['id'] = range(1, len(df) + 1)
source to share
Well, as you said, k-mean wants a vector per input, whereas you provide it with a 3D array per image. The easiest way to solve such a problem (which requires some creativity) is to develop a set of functions that can be recognized for the classes you have.
Since in this case you want to classify between nature (lot o "green"), water (lot o "blue" and "sunset" (lot o 'read / yellow / pink, perhaps?), You can use general or medium green blue and red values To check if the functions you have selected are discriminatory, you can plot a histogram.
go from your 4D array (image size x width x height) to a 2D array (image x in medium color). You need to take np.mean for different colors, height and width. At the end you should have an array (images x 3 (colors)).
source to share
This is because you are missing a 4d array if a 2d is expected. 'img_array.shape' should be like this (n_samples, n_features). You need to use a data extraction algorithm.
This can be done using the scikit-image module. You need to convert your images to grayscale format. Code:
import skimage.feature as feature
img_converted = []
for i in range(len(img_array)):
img_converted.append(feature.hog(img_array[i]))
model.fit(np.array(img_converted))
Documentation: http://scikit-image.org/docs/dev/api/skimage.feature.html#hog
source to share