The present application belongs to the technical field of
image processing, and in particular to a multi-scale attention weak light
image enhancement method and
system. The present application comprises the following steps: step 1, constructing a
network model: constructing a multi-scale attention weak light
image enhancement network model, which comprises a
decomposition network, a reflection
image restoration network, and an illumination
image enhancement network; step 2, preparing a
data set: using the LOL-v1
data set to divide the
training set and the
test set and to perform preprocessing; using the LOL-v2
data set to fine-tune the model; step 3, training the
network model: inputting the LOL-v1 data into the network for training until the preset training number or the
loss function reaches the standard step; step 4, fine-tuning the model: using the second weak light image data set to further
train and fine-tune the network model; and step 5, saving the model: solidifying and saving the final
model parameters to realize the conversion output of weak light images into high-quality images. The present application designs an efficient multi-scale attention enhancement network for improving the performance of image features. This method can significantly improve the quality and
clarity of the image, making the illumination and reflection details more distinct.