evan-person / batch-image-enhacement

Pipeline code to stitch together several image enhancement steps and process a folder of images. Offers heuristics and statistics for before/after enhancement. Currently only using openCV modules, but other modules are planned to be added.

batch-image-enhacement

Pipeline code to stitch together several image enhancement steps and process a folder of images. Offers heuristics and statistics for before/after enhancement. Currently only using openCV modules, but other modules are planned to be added.

Run process

$ runscript.py --pipeline current_pipeline.json --input_directory ~/datafolder --output_directory ~/outputfolder

Currently implemented processes:

CLAHE - contrast limited adaptive histogram equalization

Uses window to locally adjust Lightness histogram in LAB color space to adaptively improve contrast function name: clahe() parameters:

• tile_size: default is 20, sets comparison window size for localized histogram equalization
• clip_limit: default is 2.0, sets clipping limit for CLAHE

RGB equalization

Stretches histograms for R,G,B channels separately to attempt to fix color function name: rgb_eq() parameters:

• none currently, should change to allow batch norm

Gamma correction

Scales brightness to make minimum brightness above a certain level function name: gamma_corr() parameters:

• minimum_brightness: default is 0.3, sets minimum gamma
• planned to add additional inputs to allow batch norm

Unsharp Mask

Applies an unsharp mask to image to enhance edges function name: unsharp_mask() parameters:

• kernel_size: default is 5, size of gaussian blur filter used, assuming square filter of equal dimension
• sigma: default is 2.0, sigma of gaussian blur filter used
• amount: default is 2.0, amount of "sharpness" to apply
• threshold: default is 0, if positive non-zero, sets threshold for mask to use original image instead of

Median filter

Applies a median filter to image to reduce shot noise parameters:

• kernel_size: default is 7

Gaussian blur

Applies a Gaussian blur to the image parameters:

• kernel_size: default is 9, assumes square filter
• sigma: default is 0, sigma of Gaussian

Bilateral filter

Applies a bilateral filter to the image parameters:

• diameter: default is 9
• sigma_color: default is 100
• sigma_space: default is 100

Notes:

• 'params' field in pipeline json needs to have an empty dict ('{}') if you don't want to pass any arguments

Things to add:

• batch normalization
• add statistics computation
• add median, bilinear, and gaussian filters
• tests for input and modules