Content-Aware Dark Image Enhancement Through Channel Division
A.Rivera,B.Ryu & O.Chae, IEEE transactions on Image Processing,Vol.21,No.9,Sep 2012
The content aware algorithm enhances the dark image by preserving the details in textured regions and smoothness in flat regions. The content of the image is analyzed and the transformation function is determined based on the analysis of results.The content-aware method analyses the contrast in the boundaries and in texture regions.The different characteristics of the image are grouped. The groups has to be treated independently so specific functions for each group is built that will enhance the characteristics. This enhancement is due to the transformation function that was created by the analysis of the image. It preserves the image characteristics while preserving the details.
STEPS:
CONTRAST ---> INTENSITY ---> REGION PAIR CHANNEL CHANNEL
The first step is creating of contrast pairs. The contrast pairs are then accumulated into Local Contrast Indicator(LCI) functions that are merged into channels. The channels are then grouped into region channel. The region channel enhances the specific characteristics of each image and merge the results of the process to reduce the artifacts for maximum enhancement.
STEPS:
CONTRAST ---> INTENSITY ---> REGION PAIR CHANNEL CHANNEL
The first step is creating of contrast pairs. The contrast pairs are then accumulated into Local Contrast Indicator(LCI) functions that are merged into channels. The channels are then grouped into region channel. The region channel enhances the specific characteristics of each image and merge the results of the process to reduce the artifacts for maximum enhancement.
The image is transformed to hue-saturation-value color space. The proposed algorithm is applied to the intensity component. The hue and saturation components from the original image are merged with the enhanced intensity component to create a final image. The color of the image is maintained while improving the intensity component.
Contrast Pair:
Contrast is modeled through contrast pairs. It is nothing but the intensity difference between two pixels in the image.This pair acts like a force that will separate the intensities. The isolated pixels will maintain their intensities due to lack of interaction. This contrast pair is constructed from the eight neighbors of each pixel from the image. It is done by scanning the top-left neighbors of the pixel(three neighbors above the pixel and one directly to the left)and the four pairs are processed by scanning the bottom right neighbors . The contrast pairs of the image is classified into edge (boundaries and texture) and smooth(flat)regions.If it exceeds a certain threshold, then it is considered as the edge pair otherwise it is considered as the smooth pair.
LCI is nothing but the accumulation of votes by the contrast pair and they produce different slopes according to their accumulation.
Intensity Channel:
The contrast pairs are then grouped into intensity channels(LCI) that corresponds to each intensity pair. The transformations constructed from the intensity channel produce better results. The intensity channels enhances the contrast of the image without introducing artifacts.
Region Channels:
To produce best results, channels with similar characteristics are grouped into region channels. The region channel hold dark, middle and bright intensities.Each channel has a different weighting function which accentuate its characteristics. The weighting functions are then shifted to gaussian function which is normalized to one.
This method enhances image from wide variety of environments, thereby reducing the artifacts and other unnatural effects.
Contrast Pair:
Contrast is modeled through contrast pairs. It is nothing but the intensity difference between two pixels in the image.This pair acts like a force that will separate the intensities. The isolated pixels will maintain their intensities due to lack of interaction. This contrast pair is constructed from the eight neighbors of each pixel from the image. It is done by scanning the top-left neighbors of the pixel(three neighbors above the pixel and one directly to the left)and the four pairs are processed by scanning the bottom right neighbors . The contrast pairs of the image is classified into edge (boundaries and texture) and smooth(flat)regions.If it exceeds a certain threshold, then it is considered as the edge pair otherwise it is considered as the smooth pair.
LCI is nothing but the accumulation of votes by the contrast pair and they produce different slopes according to their accumulation.
Intensity Channel:
The contrast pairs are then grouped into intensity channels(LCI) that corresponds to each intensity pair. The transformations constructed from the intensity channel produce better results. The intensity channels enhances the contrast of the image without introducing artifacts.
Region Channels:
To produce best results, channels with similar characteristics are grouped into region channels. The region channel hold dark, middle and bright intensities.Each channel has a different weighting function which accentuate its characteristics. The weighting functions are then shifted to gaussian function which is normalized to one.
This method enhances image from wide variety of environments, thereby reducing the artifacts and other unnatural effects.