texton_color_utils.py

import cv2
import sys
import numpy as np
from TextonsSeg.create_LM_filters import LMFilters
from TextonsSeg.convolve_LM_filters import preprocessImageWithKernels
from TextonsSeg.get_final_vectors import createVector
from TextonsSeg.K_Means import KMeansLMfilters
from TextonsSeg.reconstruct_image import reconstructImage
import random
random.seed(128)

class Textons(object):
    def __init__(self, image, cluster_centers, iterations, type_of_assignment):
        """
        inputs:
            1.image ==> numpy array (grayscaled image)
            2.number of cluster centers ==> integer
            3.number of iterations ==> integers
            4.type of colors assignment in the final image ==> integer 0 for 'RANDOM' or 1 for'DEFINED'

        output:
            numpy array of image after k means wiht LM filters
        """
        
        try:
            self.im_color = image
            self.im_intermediate  = np.zeros_like(self.im_color)
            self.im_intermediate[:] = self.im_color.mean(axis=-1, keepdims = 1)
            self.im = self.im_intermediate[:,:,0]
            
        except:
            print("Error loading image! Not a numpy array!!")
            sys.exit()

        self.cluster_centers = cluster_centers
        self.iterations = iterations
        self.type_of_assignment = type_of_assignment

    def textons(self):
        # Criacao dos filtros descritos no paper Leung & Malik 
        LM_filters = LMFilters()
        
        # Importação desses filtros como np.array
        filters = np.array(LM_filters.makeLMfilters())
        #check filters
        #print("filters created...")
        #print(filters.shape)
        #genrate vectors applying kMeans
        # Aplicando preprocessamento
        I = preprocessImageWithKernels(self.im, self.im_color)
        
        # É realmente necessário aplicar gaussian blur na imagem?
        I.merge()
        I.apply_kernel(filters) 
        ##I.apply_kernel(np.load("LMkernels.npy"))
        features_for_kmeans, color_image = I.create_vectors()
        #print(features_for_kmeans.shape)
        ##np.save("vectors_for_kmeans.npy", features_for_kmeans)
        #print("creating first vector set..")
        #print(features_for_kmeans.shape)
        #create final vectors
        V = createVector(features_for_kmeans, self.im, color_image)
        final_feature_vectors = V.generateVectors()
        #print(final_feature_vectors.shape)
        #print("creating final vector set...")
        #print(final_feature_vectors.shape)
        #np.save("final_feature_vectors.npy", final_feature_vectors)
        ##apply k means on higher dimension image
        K = KMeansLMfilters(final_feature_vectors, no_of_clusters=self.cluster_centers,no_of_iterations=self.iterations)
        final = K.kMeans()
        #print("done")
        #reconstrcut image after k means
        final_image = reconstructImage(final,image=self.im,number_of_centers=self.cluster_centers, 
                                        assignment_type= self.type_of_assignment)
        display_image = final_image.reconstruct()
        #print('final image array ...')
        return display_image