A Deep Learning Approach for Reconstruction of Color Images in Different Lighting Conditions Based on Autoencoder Technique

Object detection problem in terms of different lighting conditions has been a challenging issue. Existing algorithms can only detect the objects with their shapes. However, when the color of an object changes in different times of a day for its various lighting conditions, these models fail to detect the shape of those color changing objects. As a result, the reliability of those models decreases. This model proposes an Autoencoder technique which can transfer an image of an object to its exact color. A new dataset is created where the image of an object is taken in two different lighting conditions to represent change in color of the same object. Then an autoencoder technique is applied on this dataset. The main function of an Autoencoder is to reconstruct its input image which is given in output through a neural network. Once the object is reconstructed to its exact color, understanding the object for any model becomes much more efficient in comparison with existing object detection models.