INTRODUCTION

When a person captures images in low-light(without light effect), the images lack visibility. multimedia algorithms are mostly designed for good quality image inputs that don't accurately satisfy low light dark images. Low–light setting is an integral part of our everyday activities. As day changes to night-time, because of darkness image clarity decreases and the captured images look darker and with more noise. HDR quality imaging can produce a higher dynamic range of brightness than it produces in standard cameras. Standard cameras take photographs with a limited exposure range, referred to as LDR, resulting in the loss of detail in highlights or shadows. Most cameras cannot provide the exposure range values within a single exposure, due to their low dynamic range. HDR photographs in general are achieved by taking different overexposure images, often by using exposure bracketing technique, and then by concatenating them into a single HDR image. To produce a good quality image Recently, more complex models such as deep convolutional neural networks (CNNs) have been used to address this problem . Here proposed CNN to check the quality of images. Image restoration plays an important role nowadays so here used the haze Removal method. The main objective of the haze removal algorithm is to enhance the image and to restore the information of the scene from a hazy image.Peoples require images to be with more clarity, so to produce an image without noise, wavelet denoising technique is used. And to enhance the image exposure fusion methodology is used. All these techniques provide an enhanced HDR tone image without noise and with good quality.

ARCHITECTURE DIAGRAM 

REQUIREMENTS
Software Requirements

●  Software : Visual studio code

●  Programming Language : Python.

●  Packages Used : numpy, scipy, Image, etc.

Hardware Requirements

●  RAM : 4GB and above (Preferably 8GB).

●  Display : HD display (1280 by 720 pixels) and above.

●  Processor : Intel i5 and above.

DATASET
● Real time dataset
● Multiple images with visibly different exposures of the same scene. The images are allowed with a subtle amount of shake.
● Images with noise can be passed as an input.
● Blurred images are allowed.
● Low light dark Images and also LDR images can be passed.

CONCLUSION

In this project, proposed a CNN algorithm to produce accurate results to check the quality of images. Introduced an image replication method to convert a single low light image to multiple exposure low Light images to reduce the noise of the image. Approached new dehazing techniques to obtain haze-free results. Proposed a Laplacian-Gaussian pyramid based fusion for multiple exposed images. This algorithm is driven by weight maps where images are converted into grayscale and using laplacian filters, absolute values of filter response are calculated. The HDR tone image is then obtained by collapsing the stack using weighted blending. Since 3 replica images with different exposure are produced, exposure fusion seamlessly gives an enhanced dark image. Band ratioing algorithm is proposed to produce a standard image by reducing the brightness values. Ratio images can also be used to generate false color composite. Proposed a wavelet transformation method to reduce the noise of the image by converting the image to grayscale. This algorithm finds many distinguishing features from even small objects. The proposed algorithm works as a single low light enhancement software as well.

WORKING VIDEO

https://drive.google.com/file/d/1C1i1OS7GyPnJHzhvs0qQNRdR_Kteaf0F/view?usp=drivesdk

RESULTS

Input image

Output image