[PyGAD] Genetic Algorithm (GA) : Image Reconstruction

PyGAD 알고리즘을 사용하여, 각 이미지의 픽셀값을 진화시켜 이미지를 재생성합니다. 진행순서는 다음과 같습니다.

1. 이미지 로드
2. Fitness function 정의
3. pygad.GA class를 활용하여 인스턴스 생성
4. PyGAD 실행
5. Plot results 확인
6. 통계 계산

1. 이미지 로드

이번 예시에서는 fruit.jpg 이미지를 활용합니다.

import numpy
import imageio
import gari
import pygad
import matplotlib.pyplot
import imageio.v2 as imageio

염색체의 표현은 픽셀값으로, 0~255의 값으로 표현합니다.

# Reading target image to be reproduced using Genetic Algorithm (GA).
target_im = imageio.imread('fruit.jpg')
target_im = numpy.asarray(target_im/255, dtype=numpy.float64)

2. Fitness function 정의

Fitness function을 정의하기 전, gari.img2chromosome() 메소드를 사용해서 이미지를 벡터로 표현시킵니다.

그 후, 각 generation별 Fitness value를 계산하기 위해 Fitness function을 정의합니다. Fitness value는 원본 이미지의 염색체와 재생성되는 이미지의 염색체의 차의 합으로 정의입니다.

# Target image after enconding. Value encoding is used.
target_chromosome = gari.img2chromosome(target_im)

def fitness_fun(ga_instance,solution, solution_idx):
    """
    Calculating the fitness value for a solution in the population.
    The fitness value is calculated using the sum of absolute difference between genes values in the original and reproduced chromosomes.

    solution: Current solution in the population to calculate its fitness.
    solution_idx: Index of the solution within the population.
    """

    fitness = numpy.sum(numpy.abs(target_chromosome-solution)) # here is the fitness function; you can modify to see if the search can be improved

    # Negating the fitness value to make it increasing rather than decreasing.
    fitness = numpy.sum(target_chromosome) - fitness
    return fitness

last_fitness = 0
def on_generation(ga_instance):
    global last_fitness

    if (ga_instance.generations_completed % 50 ==0):

      print("Generation = {generation}".format(generation=ga_instance.generations_completed))
      print("Fitness    = {fitness}".format(fitness=ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness)[1]))
      print("Change     = {change}".format(change=ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness)[1] - last_fitness))
      last_fitness = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness)[1]
      if ga_instance.generations_completed % 500 == 0:
        matplotlib.pyplot.imsave('solution_'+str(ga_instance.generations_completed)+'.png', gari.chromosome2img(ga_instance.best_solution()[0], target_im.shape))

3. pygad.GA class를 활용하여 인스턴스 생성

# The GA parameters are set
ga_instance = pygad.GA(num_generations=5000,
                       num_parents_mating=10,
                       fitness_func=fitness_fun,
                       sol_per_pop=20,
                       num_genes=target_im.size,
                       init_range_low=0.0,
                       init_range_high=1.0,

                       crossover_type="uniform",  # you could change with other crossover operations such as single_point, two_points, uniform, scattered
                       mutation_percent_genes=0.01,
                       mutation_type="random", # you could change with other mutation operations such as random, swap, inversion, scramble
                       mutation_by_replacement=True,
                       random_mutation_min_val=0.0,
                       random_mutation_max_val=1.0,
                       on_generation=on_generation,
#                        parallel_processing=["process", 5])
                       parallel_processing=5)

4. PyGAD 실행

ga_instance.run()

5. Plot results 확인

ga_instance.plot_fitness()

6. 통계 계산

최적의 솔루션에 대한 정보를 확인할 수 있습니다.

# Returning the details of the best solution.
solution, solution_fitness, solution_idx = ga_instance.best_solution()
print("Fitness value of the best solution = {solution_fitness}".format(solution_fitness=solution_fitness))
print("Index of the best solution : {solution_idx}".format(solution_idx=solution_idx))

if ga_instance.best_solution_generation != -1:
    print("Best fitness value reached after {best_solution_generation} generations.".format(best_solution_generation=ga_instance.best_solution_generation))

다음을 통해 재생성된 이미지를 확인할 수 있습니다.

result = gari.chromosome2img(solution, target_im.shape)
matplotlib.pyplot.imshow(result)
matplotlib.pyplot.title("PyGAD & GARI for Reproducing Images")
matplotlib.pyplot.show()

---

이 외에도, number of generation, crossover, mutation의 파라미터를 변경하여 다양한 실험이 가능합니다. 분명한 것은 더 많은 세대를 지날수록 이미지 재생성의 퍼포먼스는 더욱 더 향상되는 것을 확인할 수 있을 것입니다.