Underwater image enhancement method and system based on structure decomposition and storage medium
An underwater image and construction method technology, which is applied in image enhancement, image data processing, neural learning methods, etc., can solve the problems of insufficient image clarity, loss, considering the cycle consistency of input images and generated images, etc.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0092] An embodiment of the present application provides an underwater image enhancement method based on structural decomposition, including: acquiring the original first underwater image; decomposing the original first underwater image into first high-frequency information HFI and first low-frequency information LFI The UWCNN-SD network model that completes the training includes a preliminary enhanced network model, and the preliminary enhanced network model includes an LF enhanced network model and a HF enhanced network model; the first high-frequency information HFI is input into the HF enhanced network model that completes the training to Obtain the first enhanced high-frequency information after the enhancement, input the first low-frequency information LFI into the LF enhanced network model that has completed the training to obtain the first enhanced low-frequency information after the enhancement; combine the first enhanced high-frequency information after the enhancement...
Embodiment 2
[0152] Such as Figure 4 As shown, the difference from Embodiment 1 is that the UWCNN-SD network model that has been trained also includes a refined network model.
[0153] Among them, the refined network model includes seven sequentially connected convolutional layers, BN and activation functions, which are constructed as follows:
[0154] (3×3)×3×32 convolution layer (step size 1, no padding, BN, LReLU), (3×3)×32×32 convolution layer (step size 1, no padding, BN , LReLU), (3×3)×32×32 convolution layer (step size 1, no padding, BN, LReLU), (3×3)×32×32 convolution layer (step size 1, No padding, BN, LReLU), (3×3)×32×32 convolutional layer (stride 1, no padding, BN, LReLU), (3×3)×32×32 convolutional layer (step Length is 1, no padding, BN, LReLU), (1×1)×32×3 convolutional layer (step size is 1, no padding), the network structure is shown in Table 3.
[0155] table 3
[0156]
[0157]
[0158] Such as Figure 5 As shown, an underwater image enhancement method based on s...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


