色調及SVM的開關自適應中值濾波器用於鹽和胡椒去噪

Abstract

在影像處理中，影像捕捉期間的感測器故障以及數位類比轉換器的故障會導致椒鹽雜訊。椒鹽雜訊將隨機極值引入影像像素，這可能會降低影像品質。因此，我們應該在其他影像處理任務之前進行影像去噪，以確保後續操作的準確性。目前，椒鹽雜訊去噪的方法主要有三種：機器學習、變分方法和基於中位數的濾波器。機器學習需要大量訓練資料來學習影像特徵，但隱私和資料缺乏可能會成為問題。變分方法表現出優異的去噪性能，但計算複雜度高，參數選擇困難。然而，基於中位數的濾波器不需要大量的訓練數據，且計算複雜度較低。儘管如此，濾波器的降噪性能可能會受到雜訊參數的影響。因此，我們研究我們的新穎演算法。 我們提出了用於椒鹽去雜訊的「Hue-SVM Switch Adaptive Median-Based Filter」。我們對自適應中值濾波器進行調整以進行彩色影像去噪。我們利用HSI顏色模型和SVM分類器在我們提出的兩個基於中值的濾波器之間自動切換。我們的濾波器可以根據雜訊參數靈活選擇演算法。此外，與其他基於中值的濾波器相比，我們的濾波器可以實現更高品質的去噪影像。在模擬結果中，我們的濾波器在不同的雜訊參數下展現了最佳的去噪能力。

In image processing, sensor malfunctions during image capture and faults in the digital-to-analog converter result in salt-and-pepper noise. Salt-and-pepper noise introduces random extreme values into image pixels, which may decrease image quality. Therefore, we should perform image denoising before other image processing tasks to ensure the accuracy of subsequent operations. Currently, there are three main approaches for denoising salt-and-pepper noise: machine learning, variational methods, and median-based filters. Machine learning requires a lot of training data to learn image features, but privacy and a lack of data can be problems. Variational methods exhibit excellent denoising performance but come with high computational complexity and difficulty in parameter selection. However, median-based filters do not require a lot of training data and have low computational complexity. Nonetheless, the denoising performance of filters can be affected by noise parameters. Therefore, we research our novel algorithms. We propose the Hue-SVM Switch Adaptive Median-Based Filter for salt-and-pepper denoising. We make adjustments to the adaptive median filter for color image denoising. We utilize the HSI color model and SVM classifier for automatic switching between the two median-based filters that we propose. Our filter can flexibly select algorithms based on noise parameters. Moreover, our filter achieves higher-quality denoised images compared to other median-based filters. In the simulation results, our filter demonstrates the best denoising capability across different noise parameters.