Pokemon Color Transfer - Revision history

Yifdeng: /* Background */

2025-12-12T18:06:08Z

Background

← Older revision		Revision as of 18:06, 12 December 2025
Line 28:		Line 28:
	* '''Value (V)''': Brightness (0-1), from black to full intensity		* '''Value (V)''': Brightness (0-1), from black to full intensity

	'''~~Why~~ HSV for Color Transfer?'''		'''Reasons to use HSV for Color Transfer'''

	We employ HSV alongside LAB in our pipeline for complementary purposes:		We employ HSV alongside LAB in our pipeline for complementary purposes:

Yifdeng: /* Color and Style Transfer */

2025-12-12T18:05:43Z

Color and Style Transfer

← Older revision		Revision as of 18:05, 12 December 2025
Line 19:		Line 19:
	Style and color transfer [1] is a fundamental technique in computer vision and computational photography that aims to reinterpret an image by borrowing visual characteristics from another. Early work focused on texture synthesis and color statistics matching, enabling simple recoloring or pattern transfer across images. With the rise of deep learning, neural style transfer demonstrated that high-level artistic patterns—such as brushstrokes, textures, and global color mood—could be extracted from one image and blended with the structural content of another [2]. Modern approaches further incorporate perceptual features, palette manipulation, and learned representations to achieve more faithful, controllable, and efficient transformations. These methods are now widely used in artistic rendering, photo enhancement, film post-production, augmented reality, and low-power edge-device imaging pipelines [3].		Style and color transfer [1] is a fundamental technique in computer vision and computational photography that aims to reinterpret an image by borrowing visual characteristics from another. Early work focused on texture synthesis and color statistics matching, enabling simple recoloring or pattern transfer across images. With the rise of deep learning, neural style transfer demonstrated that high-level artistic patterns—such as brushstrokes, textures, and global color mood—could be extracted from one image and blended with the structural content of another [2]. Modern approaches further incorporate perceptual features, palette manipulation, and learned representations to achieve more faithful, controllable, and efficient transformations. These methods are now widely used in artistic rendering, photo enhancement, film post-production, augmented reality, and low-power edge-device imaging pipelines [3].
	The aim of color transfer in this project is to reinterpret an image by borrowing visual characteristics from another, and modify the color appearance of a pair of images so that it adopts the color style, mood, or chromatic distribution of a another image, and looks good to human.		The aim of color transfer in this project is to reinterpret an image by borrowing visual characteristics from another, and modify the color appearance of a pair of images so that it adopts the color style, mood, or chromatic distribution of a another image, and looks good to human.


	=== HSV Color Space ===		=== HSV Color Space ===

Yifdeng: /* HSV Color Space */

2025-12-12T18:05:25Z

HSV Color Space

← Older revision		Revision as of 18:05, 12 December 2025
Line 33:		Line 33:
	We employ HSV alongside LAB in our pipeline for complementary purposes:		We employ HSV alongside LAB in our pipeline for complementary purposes:

	'''Intuitive Color Shifts''': HSV allows direct manipulation of color properties. Hue differences (ΔH) naturally represent color-to-color mappings (e.g., green → blue), while saturation and value adjustments preserve tonal relationships.		* '''Intuitive Color Shifts''': HSV allows direct manipulation of color properties. Hue differences (ΔH) naturally represent color-to-color mappings (e.g., green → blue), while saturation and value adjustments preserve tonal relationships.

	'''Perceptual Separation''': By decoupling chromatic content (H, S) from luminance (V), we can modify colors while preserving shading and depth cues that define Pokémon structure.		* '''Perceptual Separation''': By decoupling chromatic content (H, S) from luminance (V), we can modify colors while preserving shading and depth cues that define Pokémon structure.

	'''Gradient Preservation''': When transferring between palette colors, HSV shifts maintain smooth color transitions better than direct RGB interpolation. This is critical for preserving gradients in Pokémon artwork (e.g., shading on Squirtle's shell).		* '''Gradient Preservation''': When transferring between palette colors, HSV shifts maintain smooth color transitions better than direct RGB interpolation. This is critical for preserving gradients in Pokémon artwork (e.g., shading on Squirtle's shell).

	'''Complementary to LAB''': While LAB excels at perceptual distance measurement (used in palette matching), HSV excels at color transformation. Our methods use LAB to determine ''which'' colors should map to each other, then HSV to compute ''how'' to shift pixel colors.		* '''Complementary to LAB''': While LAB excels at perceptual distance measurement (used in palette matching), HSV excels at color transformation. Our methods use LAB to determine ''which'' colors should map to each other, then HSV to compute ''how'' to shift pixel colors.

	'''Implementation in Our Methods''':		'''Implementation in Our Methods''':

Yifdeng: /* Color and Style Transfer */

2025-12-12T18:04:46Z

Color and Style Transfer

← Older revision		Revision as of 18:04, 12 December 2025
Line 21:		Line 21:


	==== HSV Color Space ====		=== HSV Color Space ===

	HSV (Hue-Saturation-Value) is a cylindrical color space that separates chromatic information from intensity, making it particularly suitable for color manipulation tasks. Unlike RGB, which mixes color and brightness, HSV represents:		HSV (Hue-Saturation-Value) is a cylindrical color space that separates chromatic information from intensity, making it particularly suitable for color manipulation tasks. Unlike RGB, which mixes color and brightness, HSV represents:
Line 33:		Line 33:
	We employ HSV alongside LAB in our pipeline for complementary purposes:		We employ HSV alongside LAB in our pipeline for complementary purposes:

	# '''Intuitive Color Shifts''': HSV allows direct manipulation of color properties. Hue differences (ΔH) naturally represent color-to-color mappings (e.g., green → blue), while saturation and value adjustments preserve tonal relationships.		'''Intuitive Color Shifts''': HSV allows direct manipulation of color properties. Hue differences (ΔH) naturally represent color-to-color mappings (e.g., green → blue), while saturation and value adjustments preserve tonal relationships.

	# '''Perceptual Separation''': By decoupling chromatic content (H, S) from luminance (V), we can modify colors while preserving shading and depth cues that define Pokémon structure.		'''Perceptual Separation''': By decoupling chromatic content (H, S) from luminance (V), we can modify colors while preserving shading and depth cues that define Pokémon structure.

	# '''Gradient Preservation''': When transferring between palette colors, HSV shifts maintain smooth color transitions better than direct RGB interpolation. This is critical for preserving gradients in Pokémon artwork (e.g., shading on Squirtle's shell).		'''Gradient Preservation''': When transferring between palette colors, HSV shifts maintain smooth color transitions better than direct RGB interpolation. This is critical for preserving gradients in Pokémon artwork (e.g., shading on Squirtle's shell).

	# '''Complementary to LAB''': While LAB excels at perceptual distance measurement (used in palette matching), HSV excels at color transformation. Our methods use LAB to determine ''which'' colors should map to each other, then HSV to compute ''how'' to shift pixel colors.		'''Complementary to LAB''': While LAB excels at perceptual distance measurement (used in palette matching), HSV excels at color transformation. Our methods use LAB to determine ''which'' colors should map to each other, then HSV to compute ''how'' to shift pixel colors.

	'''Implementation in Our Methods''':		'''Implementation in Our Methods''':

Yifdeng: /* Background */

2025-12-12T18:04:01Z

Background

@@ Line 19: / Line 19: @@
 Style and color transfer [1] is a fundamental technique in computer vision and computational photography that aims to reinterpret an image by borrowing visual characteristics from another. Early work focused on texture synthesis and color statistics matching, enabling simple recoloring or pattern transfer across images. With the rise of deep learning, neural style transfer demonstrated that high-level artistic patterns—such as brushstrokes, textures, and global color mood—could be extracted from one image and blended with the structural content of another [2]. Modern approaches further incorporate perceptual features, palette manipulation, and learned representations to achieve more faithful, controllable, and efficient transformations. These methods are now widely used in artistic rendering, photo enhancement, film post-production, augmented reality, and low-power edge-device imaging pipelines [3].
 The aim of color transfer in this project is to reinterpret an image by borrowing visual characteristics from another, and modify the color appearance of a pair of images so that it adopts the color style, mood, or chromatic distribution of a another image, and looks good to human.
 === Palette Extraction ===

Yifdeng: /* References */

2025-12-12T17:56:14Z

References

← Older revision		Revision as of 17:56, 12 December 2025
Line 686:		Line 686:

	[4] Ahmed, Mohiuddin, Raihan Seraj, and Syed Mohammed Shamsul Islam. "The k-means algorithm: A comprehensive survey and performance evaluation." Electronics 9.8 (2020): 1295.		[4] Ahmed, Mohiuddin, Raihan Seraj, and Syed Mohammed Shamsul Islam. "The k-means algorithm: A comprehensive survey and performance evaluation." Electronics 9.8 (2020): 1295.

			[5] Wang, Y., Liu, Y. and Xu, K. (2019), An Improved Geometric Approach for Palette-based Image Decomposition and Recoloring. Computer Graphics Forum, 38: 11-22. https://doi.org/10.1111/cgf.13812

			[6] Zheng-Jun Du, Kai-Xiang Lei, Kun Xu, Jianchao Tan, and Yotam Gingold. 2021. Video recoloring via spatial-temporal geometric palettes. ACM Trans. Graph. 40, 4, Article 150 (August 2021), 16 pages. https://doi.org/10.1145/3450626.3459675

	== Appendix I ==		== Appendix I ==

Yifdeng: /* Experiment 1 */

2025-12-07T21:55:11Z

Experiment 1

← Older revision		Revision as of 21:55, 7 December 2025
Line 428:		Line 428:
	\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| '''15.46'''/21.98 \|\| '''14.13'''/21.83 \|\| '''13.73'''/'''18.18''' \|\| '''15.76'''/22.05 \|\| '''17.29'''/'''18.31'''		\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| '''15.46'''/21.98 \|\| '''14.13'''/21.83 \|\| '''13.73'''/'''18.18''' \|\| '''15.76'''/22.05 \|\| '''17.29'''/'''18.31'''
	\|}		\|}
	\|}

	[[File:Bulbasaur_Squirtle_all_methods.png\|none\|thumb\|1200px\|Figure 1: Results of Bulbasaur and Squirtle]]		[[File:Bulbasaur_Squirtle_all_methods.png\|none\|thumb\|1200px\|Figure 1: Results of Bulbasaur and Squirtle]]

Yifdeng: /* Experiment 1 */

2025-12-07T21:53:57Z

Experiment 1

← Older revision		Revision as of 21:53, 7 December 2025
Line 418:		Line 418:
	\| Histogram-Chi2 \|\| '''2.79'''/36.67 \|\| 8.97/'''2.35''' \|\| 535.42/'''2.33''' \|\| 23.28/'''2.02''' \|\| 8.35/'''2.39''' \|\| 6.38/9.97		\| Histogram-Chi2 \|\| '''2.79'''/36.67 \|\| 8.97/'''2.35''' \|\| 535.42/'''2.33''' \|\| 23.28/'''2.02''' \|\| 8.35/'''2.39''' \|\| 6.38/9.97
	\|-		\|-
	\| CIELAB \|\| 19.54/14.45 \|\| 18.32/14.88 \|\| 13.34/10.08 \|\| 13.03/8.76 \|\| 18.33/14.63 \|\| 16.88/15.51		\| CIELAB \|\| 19.54/14.45 \|\| 18.32/14.88 \|\| 13.34/10.08 \|\| 13.03/'''8.76''' \|\| 18.33/14.63 \|\| 16.88/15.51
	\|-		\|-
	\| CIE94 \|\| 14.82/10.72 \|\| 13.44/11.13 \|\| '''8.84/7.41''' \|\| 8.78/'''6.71''' \|\| 13.45/10.98 \|\| 12.20/11.76		\| CIE94 \|\| 14.82/10.72 \|\| 13.44/11.13 \|\| '''8.84/7.41''' \|\| 8.78/'''6.71''' \|\| 13.45/10.98 \|\| 12.20/11.76

Yifdeng: /* Experiment 1 */

2025-12-07T21:53:27Z

Experiment 1

← Older revision		Revision as of 21:53, 7 December 2025
Line 412:		Line 412:
	! Metric !! Palette Dif !! Clustering !! Clustering-DS !! Convex Hull !! NS !! NS-S		! Metric !! Palette Dif !! Clustering !! Clustering-DS !! Convex Hull !! NS !! NS-S
	\|-		\|-
	\| FID \|\| 795.74/229.05 \|\| 106.22/110.64 \|\| 46.78/61.91 \|\| 96.12/66.19 \|\| 107.60/108.85 \|\| 75.15/42.77		\| FID \|\| 795.74/229.05 \|\| 106.22/110.64 \|\| '''46.78'''/61.91 \|\| 96.12/66.19 \|\| 107.60/108.85 \|\| 75.15/'''42.77'''
	\|-		\|-
	\| Histogram-Corr \|\| '''0.91/0.96''' \|\| '''0.91/0.96''' \|\| '''0.90/0.96''' \|\| '''0.87/0.96''' \|\| '''0.91/0.96''' \|\| '''0.90/0.96'''		\| Histogram-Corr \|\| '''0.91/0.96''' \|\| '''0.91/0.96''' \|\| '''0.90/0.96''' \|\| '''0.87/0.96''' \|\| '''0.91/0.96''' \|\| '''0.90/0.96'''
Line 420:		Line 420:
	\| CIELAB \|\| 19.54/14.45 \|\| 18.32/14.88 \|\| 13.34/10.08 \|\| 13.03/8.76 \|\| 18.33/14.63 \|\| 16.88/15.51		\| CIELAB \|\| 19.54/14.45 \|\| 18.32/14.88 \|\| 13.34/10.08 \|\| 13.03/8.76 \|\| 18.33/14.63 \|\| 16.88/15.51
	\|-		\|-
	\| CIE94 \|\| 14.82/10.72 \|\| 13.44/11.13 \|\| 8.84/7.41 \|\| 8.78/6.71 \|\| 13.45/10.98 \|\| 12.20/11.76		\| CIE94 \|\| 14.82/10.72 \|\| 13.44/11.13 \|\| '''8.84/7.41''' \|\| 8.78/'''6.71''' \|\| 13.45/10.98 \|\| 12.20/11.76
	\|-		\|-
	\| CIEDE2000 \|\| 13.97/10.06 \|\| 13.10/10.97 \|\| 9.20/7.83 \|\| 8.77/7.14 \|\| 13.11/10.81 \|\| 11.82/11.20		\| CIEDE2000 \|\| 13.97/10.06 \|\| 13.10/10.97 \|\| '''9.20/7.83''' \|\| '''8.77/7.14''' \|\| 13.11/10.81 \|\| 11.82/11.20
	\|-		\|-
	\| SSIM \|\| 0.82/0.90 \|\| '''0.96/0.95''' \|\| '''0.99/0.99''' \|\| '''0.98/0.99''' \|\| '''0.96/0.96''' \|\| '''0.97/0.95'''		\| SSIM \|\| 0.82/0.90 \|\| '''0.96/0.95''' \|\| '''0.99/0.99''' \|\| '''0.98/0.99''' \|\| '''0.96/0.96''' \|\| '''0.97/0.95'''
	\|-		\|-
	\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| 15.46/21.98 \|\| 14.13/21.83 \|\| 13.73/18.18 \|\| 15.76/22.05 \|\| 17.29/18.31		\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| '''15.46'''/21.98 \|\| '''14.13'''/21.83 \|\| '''13.73'''/'''18.18''' \|\| '''15.76'''/22.05 \|\| '''17.29'''/'''18.31'''
	\|}		\|}
	\|}		\|}

Yifdeng: /* Experiment 1 */

2025-12-07T21:45:56Z

Experiment 1

← Older revision		Revision as of 21:45, 7 December 2025
Line 424:		Line 424:
	\| CIEDE2000 \|\| 13.97/10.06 \|\| 13.10/10.97 \|\| 9.20/7.83 \|\| 8.77/7.14 \|\| 13.11/10.81 \|\| 11.82/11.20		\| CIEDE2000 \|\| 13.97/10.06 \|\| 13.10/10.97 \|\| 9.20/7.83 \|\| 8.77/7.14 \|\| 13.11/10.81 \|\| 11.82/11.20
	\|-		\|-
	\| SSIM \|\| 0.82/0.90 \|\| 0.96/0.95 \|\| 0.99/0.99 \|\| 0.98/0.99 \|\| 0.96/0.96 \|\| 0.97/0.95		\| SSIM \|\| 0.82/0.90 \|\| '''0.96/0.95''' \|\| '''0.99/0.99''' \|\| '''0.98/0.99''' \|\| '''0.96/0.96''' \|\| '''0.97/0.95'''
	\|-		\|-
	\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| 15.46/21.98 \|\| 14.13/21.83 \|\| 13.73/18.18 \|\| 15.76/22.05 \|\| 17.29/18.31		\| VGG Latent Space Distance \|\| 35.29/32.63 \|\| 15.46/21.98 \|\| 14.13/21.83 \|\| 13.73/18.18 \|\| 15.76/22.05 \|\| 17.29/18.31