Amber E
Painterly
Perlin Noise
Painterly Perlin Noise is a creative tool to enhance static images. This application allows users to transform static images of bodies of water into short videos with animated water. Using the deep learning model WaterNetV1, the system first identifies water regions from the input photo. A warping, generated by 3D Perlin Noise, is then applied to animate the water, creating a painterly motion.
Image Preprocessing
Histogram equalization and Gaussian blur
Original
Preprocessed
View Result
Preprocessed
Overlay
Mask
View Result
Mask Creation
Using WaterNetVI, an algorithm trained for water segmentation, the system creates a mask of the input image
​
The mask is then blurred to achieve a smoother transition between animated and static areas
​
Pixels identified as water regions are saved for further implementation
​
Perlin Noise Displacement Map
​Using three dimensional Perlin Noise, with the z-axis being a function of time, a displacement map is generated for X and Y.
Warp & Mix
Warp: Remap the original image using the displacement maps created with Perlin noise
​
Mix: Create copy of preprocessed image then only apply the warped pixels to the coordinates identified as water regions
Preprocessed
Fully Warped
Mixed
Final Touches
Specify total frames and frames per second
Output
