def main_pca(images_path): #Calling the function pre_process of frame_work, this returns some values in the following order, using those return values the training part and testing part is called by passing the required #arguments #return order : train_data_set,entire_train_data_as_list,no_of_classes,no_of_images_per_class,test_data_set,flag_for_testing # # ret1: train_data_set = list of all training images as an 2-d array # ret2: entire_train_data_as_list = list of all training images as a single list # ret3: no_of_classes = total number of classes in the given input database # ret4: test_data_set=list of training images as a 1-array ( In case user wants, can be typecasted to list in his script using this return value ) # ret5: count_of_dots_original_path : contains number of dots the original path has, this +1 gives number of dots to be replaced to get modified names # ret6: flag_for_testing=This is the flag which says about the directory directory structure. # flag_for_testing = 1 , means the directory structure is flat, means modifications of path names is necessary to extract class names # flag_for_testing = 0 , means the directory structure is hierarchical which means modifcations is not necessary for extracting class names train_data_set,entire_train_data_as_list,no_of_classes,test_data_set,count_of_dots_original_path,flag_for_testing=frame_work_v2.pre_process(images_path) ######### Calling traindb in train_database which actually does the training part and it returns some values which actually is needed during the testing phase. ####### Input Argument : train_data_set ( set of tranining images ) ##### It returns 3 values #### (1) mean_img : contains the mean of all the images, its a 1-d array/list ### (2) eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only ## (3) signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset mean_img,eigen_selected,signature_images_for_train_set=train_database.traindb(train_data_set) #### to find number of images trained per class ## Thas obviously no_of_images_per_class - 1 because one image will be taken for testing part #no_images_trained_per_class=no_of_images_per_class-1 #Uncomment following to print signature of the trained images #print "Printing the signature/co-relation matrix of the trained image #print signature_images_for_train_set ####### Uncomment the following lines when any lengths or the types of the signature variable are to be checked #print "signature type" #print type(signature_images_for_train_set) #print "signature length" #print len(signature_images_for_train_set) # Calling the testdb in test_database.py which takes in quite a number of arguments, lets explore the arguments # arg_1 : signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset ( which is return by train_database ) # arg_2 : test_data_set : contains the list of test data images which is randomly selected, one from each class # arg_3 : entire_train_data_as_list : contains entire train data set ( removed test_data_set from original input ) # arg_4 : mean_img : contains the mean of all the images, its a 1-d array/list ( which is return by train_database ) # arg_5 : eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only ( which is return by train_database ) # arg_6 : count_of_dots_original_path : contains number of dots the original path has, this +1 gives number of dots to be replaced to get modified names # arg_7 : flag_for_tesing : flag which actually says if the given directory structure is flat or hierarchy; It sets the flag if the structure is flat r=test_database.testdb(signature_images_for_train_set,test_data_set,entire_train_data_as_list,mean_img,eigen_selected,count_of_dots_original_path,flag_for_testing) print "efficiency is " print r return r
def pre_process(pathtoimages):
################################################# Declarations of all lists which are to be initialised ####################################
images_abs_names=[] # variable contains all the paths to file
total_img_vect=[] # variable to contain total images in vector form
mean_img_vect=[] # variable for containing mean of all images
sum_img_vect=[] # variable for containing sum of all images
mean_for_subtraction=[] # variable contains clones of mean used for subtracting
norm_list = [] # variable to hold all the norm values during testing phase
split_image_names=[] # variable to hold split image files to group into classes
class_names=[] #Variable to hold the class names
each_class=[] # variable to hold names of each class
entire_class=[] # variable to hold entire class
test_data_set=[] #variable for storing test images
train_data_set=[] #variable for storing train images
entire_train_data_as_list=[] #variable for storing train images as list
images_name_modified=[] # variable to hold modified images names
######################################################## End of Declarations #############################################################
######### get_files method in the module get_abs_names is called to get the absolute path names of all the images in input directory #######
src_img_dir=pathtoimages # Taking the backup of the directory path
images=lslR.get_files(src_img_dir) #returns all the absolute image names as a list
################# Uncomment following 2-lines to print all the absolut path names in the order given by get_abs_names
#print "printing absolute path file names as given by get_abs_names"
#print images
images_abs_names=images # Taking backup of absolute path names of the images
images.sort() # Sorting the image files so that images of each class are grouped together
################# Uncomment following 2-lines to print all the absolut path names in the order given by get_abs_names
#print "printing absolute path file names after sorting"
#print images
#_________________________________________________________________________________________________________________________________________#
# $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ For initialising mean_image_vect and sum_image_vect, $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$#
# $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ its required to know the dimension of each image, $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$#
# $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ so one test image is read and then all the $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ #
# $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ required values are found out. $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ #
#_________________________________________________________________________________________________________________________________________#
shape_image=Image.open(images[0]) # Any imge can be open, here we are opening 1st image
shape_image_array=numpy.asarray(shape_image) # Dimension of the image is to be known so converting to numpy array
shape=shape_image_array.shape # Getting the dimensions of the image convertedd array
################################### Uncomment following two-lines to know the dimension of the image
#print "printing shape or the dimension of the image"
#print shape
total_dimensions_per_image=shape[0]*shape[1] # Multiplying rows * columns of array to know total dimensions
################################### Uncomment following two-lines to know the total_dimensions of the image
#print "printing total dimensions of the image"
#print total_dimensions_per_image
#######################################initialising all the required values such as mean_img_vect, sum_img_vect
for i in range(total_dimensions_per_image):
mean_img_vect.append(0) # initialising with zeros
sum_img_vect.append(0) # initialising with zeros
######## Calling the function which returns the split file names
###### The function return_split_file names take one argument and returns two arguments
#### Input Arguments : images : contains the sorted list of absolute path names of all the images in the input directory
## Return values : return 1: split_image_names : Has the list of split image names
# Return Values : return 2: no_of_images : This contains the total number of input images
split_image_names,no_of_images=return_split_file_names(images)
####################################### Uncomment following 2 lines to see the split file names
#print "printing split file names"
#print split_image_names
#######################This part is to find the length of the path of each image to extract class name
single_image_to_find_length=split_image_names[0]
length_split_image_name=len(single_image_to_find_length)
################################ Code to extract the class names of the database
for i in range(no_of_images):
temp_class_name=split_image_names[i][length_split_image_name-2] #extracting class names
class_names.append(temp_class_name) # creating a list of class names
###################### Uncomment following 2 lines to know all the different classes with repetitions
#print "printing all class names"
#print class_names
set_of_class_names=set(class_names) #removing the repetitions using set so it contains only unique classes
###################### Uncomment following 2 lines to know all different classes without repetitions
#print "printing unique set of class names"
#print set_of_class_names
no_of_classes=len(set_of_class_names) #getting the count of no of classes
# If the directory structure is different (flat), some change to be done to the path names of the files.
# Every database has more then one class, this is obvious, coz if there are more then one class
# only then face recogniton on that database makes some sense
#Checking if the classes are partitioned properly.
flag_for_testing=0 # flag required to be set to 1 if in case if the directory structure is flat
if(no_of_classes<=1): # if number of classes is 1 it means that partition has not happend
flag_for_testing=1 # setting the flag indicating flat architecture
temp_str_for_checking_if_underscore_is_present=images[0]
temp_index_if_present=temp_str_for_checking_if_underscore_is_present.find('_') # to check if '_' is present find returns the position of the '_' in the string or else return -1 if not present
if(temp_index_if_present>=0): # if present
flag_for_changing_file_name=1 # set this flag to 1 which means that seperator is '_'
else: # if not present
flag_for_changing_file_name=0 # set this flag to 0 which means that seperator is '.' or any other symbol
# Modifying the images names so that it that seperator remains os.sep through out
for i in range(no_of_images):
if(flag_for_changing_file_name==0):
temp_image_name_modified=images[i].replace('.',os.sep,1)
else:
temp_image_name_modified=images[i].replace('_',os.sep)
images_name_modified.append(temp_image_name_modified)
#Uncomment following 2-lines to see the modified file names
#print "printing modified images names"
#print images_name_modified
########## To obtain the split image names
######## Calling the function which returns the split file names
###### The function return_split_file names take one argument and returns two arguments
#### Input Arguments : images : contains the sorted list of absolute path names of all the images in the input directory
## Return values : return 1: split_image_names : Has the list of split image names
# Return values : return 2: no_of_images : This contains the total number of input images
split_image_names,no_of_images=return_split_file_names(images_name_modified)
########### Uncomment following 2 lines to know the split file names
#print " printing list of split file names : "
#print split_image_names
#This part is to find the length of the path of each image to extract class name
single_image_to_find_length=split_image_names[0]
length_split_image_name=len(single_image_to_find_length)
#Code to extract the class names of the database
class_names=[] # making class_names to empty string which other wise contains some junk values
for i in range(no_of_images):
temp_class_name=split_image_names[i][length_split_image_name-2] #extracting class names '-2' because class names lies in last second position of list
class_names.append(temp_class_name) # creating a list of class names
########## Uncomment following two lines to know all the different classes with repetitions
#print "printing all class names"
#print class_names
set_of_class_names=set(class_names) #removing the repetitions using set so it contains only unique classes
# Uncomment to know all the different classes without repetitions
#print "printing unique set of class names"
#print set_of_class_names
no_of_classes=len(set_of_class_names) #getting the count of no of classes
no_of_images_per_class=no_of_images/no_of_classes #getting the count of no of images per class
# Uncomment to print the know the number of images per class
#print "number of images per class = %d " %(no_of_images_per_class)
#################arranging the input directory of images into the order of class
for i in range(no_of_classes):
each_class=[]
for j in range(no_of_images_per_class):
img_counter=i*no_of_images_per_class+j
each_class.append(images[img_counter])
entire_class.append(each_class) #contains all the images arranged according to the class
entire_class_backup=entire_class
#code to create trainset and testset
#one random image selected in one class will be added in testset and all other remaining (no_of_images_per_class) will be added to trainset
for i in range(no_of_classes):
image_no_for_test=random.random()*no_of_images_per_class
image_no_for_test=int(image_no_for_test)
test_data_set.append(entire_class[i][image_no_for_test])
temp_train=entire_class[i]
temp_train.remove(entire_class[i][image_no_for_test])
train_data_set.append(temp_train)
# Uncomment following lines in order to know the details of the train_data_set
#print "Printing type of train dataset
#print type(train_data_set)
#print "printing training data set"
#print train_data_set
# Uncomment following lines in order to know the details of the test_data_set
#print "Printing type of test dataset
#print type(test_data_set)
#print "printing test data set"
#print test_data_set
test_data_set_matrix=numpy.matrix(test_data_set) # converting to matrix to perform multiplication
train_data_set_matrix=numpy.matrix(train_data_set) # converting to matrix to perform multiplication
#############Uncomment following lines to know the shapes and details of train_data_set and test_data_set
#print "Printing test data shape
#print test_data_set_matrix.shape
#print "Printing train data shape
#print train_data_set_matrix.shape
################### we need the entire training data set as a single list
for r in range(no_of_classes):
c=0
for c in range(no_of_images_per_class-1):
entire_train_data_as_list.append(train_data_set[r][c])
######### Calling traindb in train_database which actually does the training part and it returns some values which actually is needed during the testing phase.
####### Input Argument : train_data_set ( set of tranining images )
##### It returns 3 values
#### (1) mean_img : contains the mean of all the images, its a 1-d array/list
### (2) eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only
## (3) signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset
mean_img,eigen_selected,signature_images_for_train_set=train_database.traindb(train_data_set)
#### to find number of images trained per class
## Thas obviously no_of_images_per_class - 1 because one image will be taken for testing part
no_images_trained_per_class=no_of_images_per_class-1
#Uncomment following to print signature of the trained images
#print "Printing the signature/co-relation matrix of the trained image
#print signature_images_for_train_set
####### Uncomment the following lines when any lengths or the types of the signature variable are to be checked
#print "signature type"
#print type(signature_images_for_train_set)
#print "signature length"
#print len(signature_images_for_train_set)
# Calling the testdb in test_database.py which takes in quite a number of arguments, lets explore the arguments
# arg_1 : signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset ( which is return by train_database )
# arg_2 : test_data_set : contains the list of test data images which is randomly selected, one from each class
# arg_3 : entire_train_data_as_list : contains entire train data set ( removed test_data_set from original input )
# arg_4 : mean_img : contains the mean of all the images, its a 1-d array/list ( which is return by train_database )
# arg_5 : eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only ( which is return by train_database )
# arg_6 : no_images_trained_per_class : contains number of images actually trained per class from the original dataset
# arg_7 : flag_for_tesing : flag which actually if the given directory structure is flat or hierarchy; It sets the flag if the structure is flat
r=test_database.testdb(signature_images_for_train_set,test_data_set,entire_train_data_as_list,mean_img,eigen_selected,no_images_trained_per_class,flag_for_testing)
return r
def train(pathtoimages): images_abs_names=[] # variable contains all the paths to file total_img_vect=[] # variable to contain total images in vector form mean_img_vect=[] # variable for containing mean of all images sum_img_vect=[] # variable for containing sum of all images mean_for_subtraction=[] # variable contains clones of mean used for subtracting norm_list = [] # variable to hold all the norm values during testing phase split_image_names=[] # variable to hold split image files to group into classes class_names=[] #Variable to hold the class names each_class=[] # variable to hold names of each class entire_class=[] # variable to hold entire class test_data_set=[] #variable for storing test images train_data_set=[] #variable for storing train images entire_train_data_as_list=[] #variable for storing train images as list # get_files method in the module get_abs_names is called src_img_dir=pathtoimages images=lslR.get_files(src_img_dir) images_abs_names=images #we might have to initialise mean_image_vect and sum_image_vect, #so we might required to know the dimension of each image, #so one test image is read and then all the required values are found out shape_image=Image.open(images[0]) shape_image_array=numpy.asarray(shape_image) shape=shape_image_array.shape total_dimensions_per_image=shape[0]*shape[1] #initialising all the required values such as mean_img_vect, sum_img_vect for i in range(total_dimensions_per_image): mean_img_vect.append(0) sum_img_vect.append(0) #the code which actually partitions the entire database of images into trainset and testset goes here no_of_images=len(images) for i in range(no_of_images): temp_image=images[i].split(os.sep) split_image_names.append(temp_image) #this part is to find the length of the path of each image to extract class name single_image_to_find_length=split_image_names[0] length_split_image_name=len(single_image_to_find_length) #Code to extract the class names of the database for i in range(no_of_images): temp_class_name=split_image_names[i][length_split_image_name-2] #extracting class names class_names.append(temp_class_name) # creating a list of class names # Uncomment to know all the different classes with repetitions #print "printing all class names" #print class_names set_of_class_names=set(class_names) #removing the repetitions using set so it contains only unique classes # Uncomment to know all the different classes without repetitions #print "printing unique set of class names" #print set_of_class_names no_of_classes=len(set_of_class_names) #getting the count of no of classes no_of_images_per_class=no_of_images/no_of_classes #getting the count of no of images per class # Uncomment to print the know thw number of images per class #print "number of images per class = %d " %(no_of_images_per_class) #arranging the input directory of images into the order of class for i in range(no_of_classes): each_class=[] for j in range(no_of_images_per_class): img_counter=i*no_of_images_per_class+j each_class.append(images[img_counter]) entire_class.append(each_class) #contains all the images arranged according to the class entire_class_backup=entire_class #code to create trainset and testset #one random image selected in one class will be added in testset and all other remaining (no_of_images_per_class) will be added to trainset for i in range(no_of_classes): image_no_for_test=random.random()*no_of_images_per_class image_no_for_test=int(image_no_for_test) test_data_set.append(entire_class[i][image_no_for_test]) temp_train=entire_class[i] temp_train.remove(entire_class[i][image_no_for_test]) train_data_set.append(temp_train) # Uncomment following lines in order to know the details of the train_data_set #print type(train_data_set) #print "printing training data set" #print train_data_set test_data_set_matrix=numpy.matrix(test_data_set) train_data_set_matrix=numpy.matrix(train_data_set) #print test_data_set_matrix.shape #print train_data_set_matrix.shape #print "printing one individual image in training data set" #print train_data_set # we need the entire training data set as a single list for r in range(no_of_classes): c=0 for c in range(no_of_images_per_class-1): entire_train_data_as_list.append(train_data_set[r][c]) # Calling traindb in train_database which actually does the training part and it returns some which actually is needed during the testing phase # It returns 3 values # (1) mean_img : contains the mean of all the images, its a 1-d array/list # (2) eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only # (3) signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset mean_img,eigen_selected,signature_images_for_train_set=train_database.traindb(train_data_set) # to find number of images trained per class no_images_trained_per_class=no_of_images_per_class-1 #print signature_images_for_train_set # Uncomment the following lines when any lengths or the types of the following variables are to be checked # print "signature type" # print type(signature_images_for_train_set) # print "signature length" # print len(signature_images_for_train_set) # Calling the testdb in test_database.py which takes in quite a number of arguments, lets explore the arguments # arg_1 : signature_images_for_train_set : contains the signatures (mapped images / eigen images ) for the entire training dataset ( which is return by train_database ) # arg_2 : test_data_set : contains the list of test data images which is randomly selected, one from each class # arg_3 : entire_train_data_as_list : contains entire train data set ( removed test_data_set from original input ) # arg_4 : mean_img : contains the mean of all the images, its a 1-d array/list ( which is return by train_database ) # arg_5 : eigen_selected : Usually only the major values of the eigen vector are taken, this contain those major eigen values only ( which is return by train_database ) # arg_6 : no_images_trained_per_class : contains number of images actually trained per class from the original dataset test_database.testdb(signature_images_for_train_set,test_data_set,entire_train_data_as_list,mean_img,eigen_selected,no_images_trained_per_class)
