| Digital images are important carriers for people to obtain visual information and perceive the outside world.With the development of science and technology,digital images are also widely used in medical detection,disaster warning and other engineering fields.However,the captured images often suffer from uneven brightness,low contrast,and noise interference due to complex lighting conditions such as night and shadow places.It is not easy to visually obtain useful information from these images,and the performance of computer vision tasks such as target detection and recognition will be affected by them.In order to improve the quality of low-light images and restore the information in them,low-light image enhancement methods are studied and the main contributions are as follows:1)A low-light image enhancement method based on normal-light image degradation is proposed.The degraded images generated from normal-light images by gamma transformation are adopted as the reference images in network training process.The brightness and contrast of the degraded images are closer to those of low-light images,which helps to improve the sensitivity of the network to details.In addition,a feature interaction network is designed to connect two encoding-decoding subnetworks in parallel.The network completes repeated feature interaction by exchanging the information across the parallel subnetworks during training.The final enhanced images are obtained by performing inverse gamma transformation on the output of the network.The experimental results show that the proposed method could effectively improve the brightness and contrast of low-light images while preserving details with good comprehensive performance,and the accuracy of object recognition task could be improved by using this method for image enhancement.2)In order to recover the color of the low-light images better,the feature interaction in the network is further increased and a low-light image enhancement method based on multi-path interaction is proposed.The method consists of three steps: color channel enhancement,reconstruction and detail adjustment.In the color channel enhancement step,the parallel encoding-decoding sub-networks are extended to form a multi-path interaction network,which enhances the R,G and B channels respectively.Information is exchanged between different paths through feature interaction,while a high-resolution channel is preserved to enrich the feature representation.In the reconstruction step,the enhanced R,G and B channels are concatenated to generate the color image.To avoid the unnatural result that may be brought by training the three channels separately,detail correction is carried out on the color image in the detail adjustment step to obtain the final enhanced image.The experimental results show that the proposed method could improve the quality of low-light images as well as effectively suppress brightness distortion and over-enhancement,the output is vivid and natural,and the image enhancement using this method is helpful to improve the performance of text recognition task. |
