This is my toy project. Explained how to draw an image with genetic algorithm and see how compress the image.

1. IMAGE GENERATION AND
COMPRESSION USING
GENETIC ALGORITHM
Kiyoung Moon

2. GENETIC ALGORITHM : GA
• Image generation using genetic algorithm aka GA

3. WHAT IS GENETIC ALGORITHM?
• In computer science and operations research, a genetic algorithm (GA) is a
metaheuristic inspired by the process of natural selection that belongs to the larger
class of evolutionary algorithms (EA). Genetic algorithms are commonly used to
generate high-quality solutions to optimization and search problems by relying on
bio-inspired operators such as mutation, crossover and selection –Wikipedia-

4. EXAMPLE OF GENETIC ALGORITHMS
• Swing, Pathfinding, Image generation, compression and so on.

5. DEMO : SWING
• It’s difficult to create an algorithm to get the most efficient ride on the swing.
• What people can do,
• Bending, Standing up and so on.
• What’s the most efficient behavior?
• To reach the highest position!
• Do not use external forces(no hack!)

6. • https://www.youtube.com/watch?v=Yr_nRnqeDp0

7. HOW GENETIC ALGORITHM WORKS
Apply Selection
Crossover/Mutation
Evaluate Fitness of
Each Individual
Create
Initial
Population
End

8. MAKE DNA
• 0 Bending
• 1 Stand up
• 2 Bounce
• 0102102010210201201021020
• 010
• Bending, Stand up, Bending
• 210
• Bounce, Stand up, Bending
• … continue

9. USE GA!
• 10201021020120 -> Person1
• 01020120101021 -> Person2
• 10201202010120 -> Person3
• 10201020201010 -> Person4
• …
• 10102020001021 -> PersonN
Who is the best person?

10. GET THE SCORE

13. MATING, BREEDING
• In the same way as real nature
• What if pretty woman, handsome man are married and get the baby,
• You can have pretty or handsome baby(well… not 100%)
• What if smartest woman, man are married and get the baby,
• Maybe he/she will be smart…(well… not 100%)
• What if most skilled people who is good at swing are married and get the baby,
• There are many different mating algorithms.
• 50% DNA from father and 50% DNA from mother…
• There are many different algorithms for this particular jobs.

14. OK, NOW WE HAVE SECOND GENERATIONS.
• Randomly crossed among N groups to make N children.
• This is the first generation.
• After that, make them to swing!
• Based on theirs's score, crossover, mutation, breeding and prepare the next
generations.
• If we continue this steps N times, we can say that the generation is changed N times.

15. IN REAL LIFE?
• There is a mutation.

16. GIVE MUTATION TO DNA
• Change order
• 010102
• 010102
• 020101
• Change value
• 010102
• 010102
• 012102

17. DO NOT HAVE THE ILLUSION OF BEING ABLE
TO BECOME AN X MAN.
• In reality… mutation is not good.

18. RARELY…

19. EVEN THOUGH YOU ARE SAME AGE,
GENERATION IS NOT SAME.

20. LET’S THINK FOR A SECOND
• What does it take to have a child?
• Spouse
• To meet spouse,
• Pretty
• Handsome
• Lots of money
• Good heart
• Smart
• Cooking skill
• Random…
• And so on

21. THINK MORE
• Would early mankind have an eye?
• Would early mankind have a nose?
• If you are good at fighting, is there better chance to get marry?
• Do people need legs in the future? Do we really need legs?

22. WHAT WE CAN DO WITH GENETIC
ALGORITHM?
• Image generation
• Pathfinding
• Creative Gameplay
• Compression
• etc.

23. COMPRESSION?
• Image is consist of pixels.

24. NUMBER OF PIXELS REQUIRED TO
REPRESENT AN IMAGE
• Width * Height * Byte per pixel
• For example
• 512 x 512 x 4 = 1048576 bytes needs.

26. DIFFERENT WAY TO REPRESENT AN IMAGE
• To draw circle,
• x, y, radius, color needs.
• 4 elements.
• To draw rectangle,
• x, y, width, height, color needs
• 5 elements.

27. • Assume one property consume 4 bytes,
• To draw one circle, we need 16 bytes
• 4 x 4 = 16 bytes
• To draw one rectangle, we need 20 bytes
• 4 x 5 = 20 bytes
• There are 2 rectangles so we need 40 bytes.
• Total bytes needs : 56 bytes
• It is 1800 times better!

28. WELL… 1800 TIMES BETTER IS LIE

30. JUST DRAW, EVALUATE SCORES
• Just draw
• How? Randomly!

31. HOW TO DRAW?
• Randomly choose x, y coordinates.
• Randomly choose color.
• Randomly choose radius.
• Randomly choose move.
• Randomly choose add more shapes.
• Randomly choose remove shapes.
• Draw

32. EVALUATE SCORE(FITNESS)
• Let original image is o and DNA’s image is p.
𝐹𝑖𝑡𝑛𝑒𝑠𝑠 =
𝑦
ℎ
𝑥
𝑤
𝑝 𝑥,𝑦 − 𝑜 𝑥,𝑦
• Low 𝐹𝑖𝑡𝑛𝑒𝑠𝑠 means good.
• Higher fitness means, difference between two images are high.
• Of course there are many different ways to compare two images.
• For instance, we can use histogram of the images.

33. OK, PEOPLE. NOW DRAW IMAGES.

35. CHOOSE BEST IMAGE
• What’s the best image?
• Lowest fitness image
• Difference between two image are low which means two images are similar!
• Crossover, Mutation
• My program does not use crossover, only used mutation.

36. COMPARE TWO IMAGES.
• If new image’s (using different DNA) fitness is lower than previous DNA image, it
means new image is more similar than previous image.
• If so update current DNA

37. Repeat this process infinitely…

39. I PAINTED IT RANDOMLY… RESULT?

40. IT WORKS

41. • After 19577 generations
• Right image is consist of 163
rectangles.

42. • After 2877684 generations,
• Used 723 rectangles.
• 723 * 4 * 4 = 11568 bytes

43. ADVANTAGES
• We can represent an image with lower space
• Fun

44. DISADVANTAGE
• To get a similar image, it took lot of time.
• There is no guarantee that the exact same image will come out.
• If we do it infinitely?

45. THINK ABOUT : CAN WE APPLY IT TO
GAMEPLAY MECHANISM?
Producer : Wow… this is cool. What we can make with this technology?
Programmer : Hmm… that is you should let me know…

46. THANK YOU
Question?