單圖像除雨方法及其系統

專利類型

發明

專利國別 (專利申請國家)

中華民國

專利申請案號

110132331

專利證號

I 780884

專利獲證名稱

單圖像除雨方法及其系統

專利所屬機關 (申請機關)

國立中正大學

獲證日期

2022/10/11

技術說明

本發明提供一種單圖像除雨方法及其系統，其可有效地去除單張圖像雨痕並恢復乾淨圖像，其中本發明之技術特徵包含以下步驟： Step 1：利用穩定小波轉換(Stationary Wavelet Transform，SWT)將1張雨圖分解為具有相同小波係數的4個子圖像(如下第1圖)。接著，將SWT分解後的各level的低頻雨圖均輸入網路除雨。 Step 2：透過wavelet approximation-attention residual (WAAR) network。WAAR network可以有效去除雨圖中各level的高頻雨紋以及低頻雨紋。 Step 3：提供一高低頻區塊連接之深度網路(如下列第3圖)，其透過高低頻區塊連接可以提取更豐富的圖像特徵，且有利於得到高低頻特徵之間的潛在相依性，進而幫助網路在雨圖高頻部分更有效地去除雨紋。 The present invention provides a single image rain removal method and system, which can effectively remove rain marks on a single image and restore a clean image. The technical features of the present invention include the following steps: Step 1: Using Stationary Wavelet Transform (SWT) to decompose a rain image into 4 sub-images with the same wavelet coefficients. Then, input the low-frequency rain map of each level after SWT decomposition into the network to remove rain. Step 2: Wavelet approximation-attention residual (WAAR) network can effectively remove high-frequency rain patterns and low-frequency rain patterns at each level in the rain map. Step 3: Providing a deep network connected by high and low frequency blocks, which can extract richer image features through high and low frequency block connections, and help to obtain the potential dependence between high and low frequency features In turn, it helps the network to more effectively remove rain patterns in the high-frequency part of the rain map.

備註

連絡單位 (專責單位/部門名稱)

技術移轉授權中心

連絡電話

05-2720411轉16001

網址