Contrast in Haze Removal Configurable Contrast Enhancement Model Based on Dark Channel Prior
Abstract: Conventional haze-removal methods are designed to adjust the contrast and saturation, and in so doing enhance the quality of the reconstructed image. Unfortunately, the removal of haze in this manner can shift the luminance away from its ideal value. In other words, haze removal involves a trade off between luminance and contrast. We reformulated the problem of haze removal as a luminance reconstruction scheme, in which an energy term is used to achieve a favorable trade off between luminance and contrast. The proposed method bases the luminance values for the reconstructed image on statistical analysis of hazefree images, thereby achieving contrast values superior to those obtained using other methods for a given brightness level. We also developed a novel module for the estimation of atmospheric light using the color constancy method. This module was shown to outperform existing methods, particularly when noise is taken into account. The proposed framework requires only 0.55 seconds to process a 1megapixel image. Experimental results demonstrate that the proposed hazeremoval framework conforms to our theory of contrast.
Existing system: Unfortunately, most existing haze-removal schemes tend to augment saturation as a direct consequence of boosting contrast. Thus, any attempt to reduce the effects of saturation augmentation work counter to attempts to boost contrast. Third, the performance of existing algorithms is usually limited due variations in haze conditions (dense or light). Furthermore, the fact that haze-removal algorithms are used for various purposes means that reconstruction accuracy and the contrast performance are worthy of consideration. These tests were meant to evaluate the lower bound execution time of constructing transmission using existing methods. Some methods can be implemented quickly; however, utilizing these methods to construct transmission maps results in visible artifacts. Proposed system:
The proposed transmission module can compete against all of the methods in the construction of transmission maps using joint-edge-preserving algorithms. The execution time of our framework is presented in Table shows the enhancement results using various setups. This experiment demonstrates that the proposed method can be construed as a generic enhancement method applicable to all types of image. The comparison in shows the reconstructed result using several methods. Results using the propose.considered a type of generic image enhancement method. The proposed method was proved highly effective when applied to images with heavy haze as well as those with fair visibility. The proposed method was evaluated using subjective assessment as well as objective measures. The proposed method is also computationally light, compared to existing methods. We posit that this work could also be extended to image matting or alpha blending due to the fact that those models are similar to atmospheric models. Advantages: The CL method is used to formulate an atmospheric light and transmission map using cross validation, in which many patches indicative of the same atmospheric light would be an indication that the estimation is true. Fattal recently proposed a
method to refine a preliminary transmission map using the Gaussian Markov Random Field (GMRF). Some researchers have used hyper-heuristic frameworks to obtain prior knowledge. Cai et al. Discovered that some trained receptive fields could be used to produce results similar to those obtained using heuristic priors, such as the dark channel prior. This is an indication that deep convolution neural networks could also be used to construct prior knowledge. In , Ren et al. formulated training sets based on the NYU depth dataset to improve the accuracy of simulations dealing with haze formation. The conventional framework used to estimate a transmission map involves two steps: The first step involves finding specific information based on heuristics. According to our experiments (Section V), priors that take into account local properties outperform those using global properties, particularly when it comes to contrast. Disadvantages: This changes the problem to one of rebuilding the entire transmission map using a limited number of reliable pixel values. Since the form of the blending atmospheric model is similar to alpha blending, the Dark Channel Prior (DCP) was implemented with the Matting Laplacian Matrix (MLM) to refine the preliminary Darkest Channel formed using a sliding minima filter. This algorithm was first propose. Any improvement in quality following the removal of haze tends to shift the luminance away from the ideal level. Over-saturation is a common problem undermining the natural appearance of images. Unfortunately, most existing hazeremoval schemes tend to augment saturation as a direct consequence of boosting contrast. Our innovation simplifies haze-removal to a problem of luminance reconstruction, wherein the energy term is adjusted to achieve a favorable tradeoff between luminance and contrast.
Once the transform A has been determined, the energy term is given a close-form solution in the Fourier domain using Parseval's theorem and the law of convolution. However, experiments revealed that this method poses problems with regard to edge alignment and dealing with HALOs around strong edges.
Modules: Image enhancement : The methods above require additional equipment or multiple images. In contrast, single image de-hazing methods focus more on image enhancement than a restoration based on strict physical laws. Most of these methods are based on augmenting the attenuated signal with a priori knowledge related to dynamic range, saturation, or contrast. Based on the observation that atmospheric light reduces contrast, Tan et al. sought to restore images by maximizing in-patch contrast defined as the sum of the gradient using an energy maximization framework. Tarel et al. used the minimal response of color channels as an alternative to the local contrast energy term. This approach proved highly effective with regard to execution time with little effect on the final results. Contract enhancement : Thus, we sought to demonstrate how a transmission map affects contrast and why conventional methods are ineffective in contrast enhancement. Experiments in Section V.B prove that our energy term makes it possible to achieve a more favorable tradeoff between luminance and contrast. The efficacy of adjusting saturation contrast is demonstrated in Section V.A. Experiments results demonstrate that decoupling color from luminance using the proposed parameter can help to reduce the loss of contrast loss in luminance reconstruction. This gives us a generic enhancement method applicable to all images, regardless of whether they are hazy, the effectiveness. Interprets saturation augmentation, and the last term should be correlated with a rational distribution of saturation. However, this interpretation does not indicate the relationship between the transmission map and contrast enhancement performance. In the following, we demonstrate that this actually works against contrast enhancement.
Color de - correlating: Using this model, a color image with 3*N known pixels would have 4*N+3 unknown variables, such that the equation set would be underdetermined. Further assumptions or prior knowledge related to scattering would be necessary to formulate an accurate prediction. Schechner et al. sought to eliminate atmospheric effects by using a polarizer to identify the location with the highest polarization, which is indicative of the maximum scattering. Kopf et al. paired a known Georeferenced Digital Terrain with an input photograph to obtain depth information directly. The objective is to identify a curve matching the true geometry of the scene in order to make parameter estimations in accordance with causal features within an input image. Gibson et al. proposed a model for the removal of haze from video clips obtained using a fixed camera position, wherein some of the unknown variables can be eliminated using a clip containing time series data. The methods above require additional equipment or multiple images. In contrast, single image de-hazing methods focus more on image enhancement than a restoration based on strict physical laws. Most of these methods are based on augmenting the attenuated signal with a priori knowledge related to dynamic range, saturation, or contrast. Haze : Haze-removal algorithms are used to obtain clean, haze-free images with enhanced saturation and contrast. Haze or fog is caused by microscopic aerosols distributed in the air. Cameras and the human eye lack the sensitivity to make out these aerosols; however, airborne particles can affect radiance in other ways, such as Rayleigh scattering, Mie scattering, and the Tyndall effect . Researchers have observed that these effects obey Koschmieder’s Law . The strength of these effects on global atmospheric light is exponentially correlated to the depth of the scene. This observation provides a novel opportunity by which to formulate estimations of atmospheric light. The blending atmospheric model proposed in accurately describes situations involving single scattering atmospheric lighting. The observed scene could be regarded as a global light veil blended with the real scene radiance at a particular ratio referred to as the transmission level. Using this model, a color image with 3*N known pixels would have 4*N+3 unknown variables, such that
the equation set would be underdetermined. Further assumptions or prior knowledge related to scattering would be necessary to formulate an accurate prediction.