MakeFeatures.py

import numpy as np
import cv2
from sklearn.cluster import MiniBatchKMeans
import pickle
import cupy as cp
import cyvlfeat as cy

dict_name= {0:'Jaffe',1:'Fer2013',2:'CK+',3:'BigFer2013'}
Fer_error_imgs = [2172, 5275, 6459, 7630, 10424, 11287, 13149, 13403, 13989, 15895, 22199, 22928, 28602, 30003]



def Extractor(Images,method_detector="FAST",data_code=1):

    Detector = None
    Descriptor = None

    if(method_detector=="FAST"):
        Detector = cv2.FastFeatureDetector.create()
        Descriptor = cv2.xfeatures2d_SIFT.create()
    else:
        Detector = cv2.xfeatures2d_SIFT.create()
        Descriptor= cv2.xfeatures2d_SIFT.create()


    desc_seq = []
    count = 0
    for img in Images:

        kp = Detector.detect(img)
        kp,desc = Descriptor.compute(img,kp)

        if(len(kp)==0):
            count +=1
            continue
        print("Image Number "+str(count)+" has been extracted !")
        desc_seq.append(desc)
        count+=1

    print("Images Extracted !")
    print("Start concatnating !")
    descriptors_data = cp.array(desc_seq[0])
    for remaining in desc_seq[1:]:
        descriptors_data = cp.vstack((descriptors_data , remaining))
    ##print("Descriptors shape :" + str(descriptors_data.shape))
    descriptors_data = cp.asnumpy(descriptors_data)
    print("End concatnating !")
    print("Descritors ready to cluster !")
    filename = dict_name[data_code]+"_"+method_detector+"feature_SIFTDescriptors.npy"

    np.save(filename,descriptors_data)
    print(filename+" has been saved to disk !")



def Vocabularyize(X_filename,data_name_code,K=2048,detector_name="SITF"):

    X = np.load(X_filename)
    print("Clustering !!!!")
    K_model = MiniBatchKMeans(n_clusters=K,max_iter=300,batch_size=K*2,max_no_improvement=30,init_size=3*K).fit(X)
    print("Clustered !!!!")
    # save the model to disk
    filename = dict_name[data_name_code]+"_"+detector_name+"Detector_Kmean_model.sav"
    pickle.dump(K_model, open(filename, 'wb'))
    print("Model Saved !!!!")


def Vetorize_of_An_Image(img,Kmean,detector_method="SIFT"):

    if detector_method == "SIFT":
        Detector = cv2.xfeatures2d_SIFT.create() ## cv2.FastFeatureDetector.create()
        Descriptor = cv2.xfeatures2d_SIFT.create()
    else:
        Detector = cv2.FastFeatureDetector.create()
        Descriptor = cv2.xfeatures2d_SIFT.create()

    vector_2048 = np.zeros(2048,dtype='uint8')
    kp = Detector.detect(img)
    kp, desc = Descriptor.compute(img, kp)

    if(len(kp)==0):
        return vector_2048

    predictions = Kmean.predict(desc)

    for pre in predictions:
        vector_2048[pre] +=1

    return vector_2048
def Histogram_All_Images(imgs,Kmean,detector_method="SIFT",data_name_code=1,):

    Stack = Vetorize_of_An_Image(imgs[0], Kmean)
    count = 0
    print("Imgage Number " + str(count) + " in Stack !")
    for img in imgs[1:]:

        vector = Vetorize_of_An_Image(img,Kmean,detector_method)
        Stack = np.vstack((Stack,vector))
        count +=1
        print("Imgage Number "+str(count)+" in Stack !")
    print("Histogram Generated ! ")
    filename = dict_name[data_name_code]+"_"+detector_method+"Detector_Histogram.npy"

    np.save(filename,Stack)
    print("Saved Histogram as numpy array to disk  !")


def Dense_SIFT_Extractor(images,data_name_code):
    kp, desc0 = cy.sift.dsift(images[0], step=1, size=1, bounds=None, window_size=1, norm=True,
                            fast=True, float_descriptors=True, geometry=(4, 4, 8),
                            verbose=False)

    count = 0
    print("Imgage Number " + str(count) + " in Stack !")
    kp, desc1 = cy.sift.dsift(images[1], step=1, size=1, bounds=None, window_size=1, norm=True,
                              fast=True, float_descriptors=True, geometry=(4, 4, 8),
                              verbose=False)

    count = 1
    print("Imgage Number " + str(count) + " in Stack !")

    concate = np.concatenate((desc0,desc1))
    for img in images[2:]:

        kp , desc = cy.sift.dsift(img, step=1, size=1, bounds=None, window_size=1, norm=True,
                              fast=True, float_descriptors=True, geometry=(4, 4, 8),
                              verbose=False)

        concate = np.concatenate((concate,desc))
        print(concate.shape)
        count +=1
        print("Imgage Number "+str(count)+" in Stack !")
    print("All Dense SIFT Descriptors Generated ! ")

    filename = dict_name[data_name_code]+"_DenseSIFF_Descriptors.npy"

    concate = np.reshape(concate,(images.shape[0],2025,128))
    np.save(filename,concate)

    print("Saved all dense sift descriptors as numpy array to disk  !")

#X = np.load("Fer_X.npy")
#X = X.astype("uint8")

#Dense_SIFT_Extractor(X,1)