def compress(self):
image_handler = ImageHandler()
kn_rows = ceil(log(self.compressorConfig.number_of_neurons, 2))
kn_columns = self.compressorConfig.number_of_neurons / kn_rows
is_RGB = image_handler.config.RGB
if is_RGB:
image_vectors_R = image_handler.split_image_to_blocks_RGB(2)
image_vectors_G = image_handler.split_image_to_blocks_RGB(1)
image_vectors_B = image_handler.split_image_to_blocks_RGB(0)
number_of_image_vectors = image_vectors_R.shape[0]
kn = KohonnenNetwork(self.compressorConfig.number_of_neurons,
image_handler.vector_dimension,
self.compressorConfig.epochs,
number_of_image_vectors,
self.compressorConfig.initial_learning_rate,
max(kn_rows, kn_columns) / 2)
random_train_vectors_R = self.drawTrainingVectors(image_vectors_R)
random_train_vectors_G = self.drawTrainingVectors(image_vectors_G)
random_train_vectors_B = self.drawTrainingVectors(image_vectors_B)
reconstruction_values_R = kn.train(random_train_vectors_R)
reconstruction_values_G = kn.train(random_train_vectors_G)
reconstruction_values_B = kn.train(random_train_vectors_B)
image_vector_indices_R, distance1 = vq(image_vectors_R,
reconstruction_values_R)
image_vector_indices_G, distance2 = vq(image_vectors_G,
reconstruction_values_G)
image_vector_indices_B, distance3 = vq(image_vectors_B,
reconstruction_values_B)
image_after_compression_RGB = np.zeros(
[image_handler.image_width, image_handler.image_height, 3],
"uint8")
for index, image_vector in enumerate(image_vectors_R):
start_row = int(index / (image_handler.image_width /
self.imageHandlerConfig.block_width)
) * self.imageHandlerConfig.block_height
end_row = start_row + self.imageHandlerConfig.block_height
start_column = (index * self.imageHandlerConfig.block_width
) % image_handler.image_width
end_column = start_column + self.imageHandlerConfig.block_width
for x in range(start_row, end_row, 1):
counter = 0
for y in range(start_column, end_column, 1):
image_after_compression_RGB[x][y][0] = \
reconstruction_values_R[image_vector_indices_R[index]][counter]
image_after_compression_RGB[x][y][1] = \
reconstruction_values_G[image_vector_indices_G[index]][counter]
image_after_compression_RGB[x][y][2] = \
reconstruction_values_B[image_vector_indices_B[index]][counter]
counter = counter + 1
output_image_name = "RGB_CB_neurons=" + str(self.compressorConfig.number_of_neurons) + "_block_size=" + \
str(self.imageHandlerConfig.block_width) + "x" + str(self.imageHandlerConfig.block_height) + ".png"
imageio.imwrite(output_image_name, image_after_compression_RGB)
mse = image_handler.mse(image_after_compression_RGB)
print("Mean Square Error = ", mse)
print("PSNR measure = ", image_handler.psnr(mse))
print("Compression ratio = ",
image_handler.compression_ratio(output_image_name))
else:
image_vectors = image_handler.split_image_to_blocks()
random_train_vectors = self.drawTrainingVectors(image_vectors)
number_of_image_vectors = image_vectors.shape[0]
kn = KohonnenNetwork(self.compressorConfig.number_of_neurons,
image_handler.vector_dimension,
self.compressorConfig.epochs,
number_of_image_vectors,
self.compressorConfig.initial_learning_rate,
max(kn_rows, kn_columns) / 2)
reconstruction_values = kn.train(random_train_vectors)
image_vector_indices, distance = vq(image_vectors,
reconstruction_values)
image_after_compression = np.zeros(
[image_handler.image_width, image_handler.image_height],
"uint8")
for index, image_vector in enumerate(image_vectors):
start_row = int(index / (image_handler.image_width /
self.imageHandlerConfig.block_width)
) * self.imageHandlerConfig.block_height
end_row = start_row + self.imageHandlerConfig.block_height
start_column = (index * self.imageHandlerConfig.block_width
) % image_handler.image_width
end_column = start_column + self.imageHandlerConfig.block_width
image_after_compression[start_row:end_row, start_column:end_column] = \
np.reshape(reconstruction_values[image_vector_indices[index]],
(self.imageHandlerConfig.block_width, self.imageHandlerConfig.block_height))
output_image_name = "CB_neurons=" + str(self.compressorConfig.number_of_neurons) + "_block_size=" + \
str(self.imageHandlerConfig.block_width) + "x" + str(self.imageHandlerConfig.block_height) + ".png"
image_handler.save_image(image_after_compression,
output_image_name)
mse = image_handler.mse(image_after_compression)
print("Mean Square Error = ", mse)
print("PSNR measure = ", image_handler.psnr(mse))
print("Compression ratio = ",
image_handler.compression_ratio(output_image_name))
