影像超解析度技術的目標,是由輸入的低解析度影像,產生出含有銳利邊緣和豐富細節材質的高品質高解析度影像。 因為一張低解析度影像,有可能由多張不同的高解析度影像產生,所以這是一個困難且不易解決的不適定問題,並且需要充足的預備知識來還原出高品質的高解析度影像。 為了解決這個問題,基於字典學習法的單一影像超解析度技術演算法,透過了學習來自數以萬計外部圖塊的對應關係,而達到了出色且先進的成果和表現。 我們提出了一個不需要使用幀緩衝器的基於字典學習法的即時超解析度系統架構。以及提出了針對此架構的改良演算法和改進架構,並且降低了面積和記憶體的用量成本。 該系統通過了TSMC 90奈米技術的驗證,並且在FPGA系統上實現,可達到每秒60張1920x1080的full HD解析度的影像。這個系統以148.36MHz的時脈運作。 我們所提出的無需使用幀緩衝器的基於字典學習法的即時超解析度系統架構,僅使用了數條行緩衝器即可由低解析度的輸入影像,產生出高解析度影像。 這個即時且無需幀緩衝器的系統,使得影像超解析度技術得以應用在影像感測器和顯示驅動系統當中,以實現計算攝影學的應用。
Image super-resolution aims to generate good quality high-resolution images with sharp edges and rich details information from input low-resolution images. It is a well-known ill-posed problem since a single low-resolution image could be generated from more than one high-resolution images, and it requires enough prior knowledge to reconstruct the high-quality high-resolution images. To solve this problem, dictionary learning-based single image super-resolution methods have achieved outstanding performance and gained state-of-the-art results, by learning information from millions of external image patches. We proposed a framebuffer-free real-time dictionary learning-based super-resolution system, as well as improved algorithm and architecture to reduce area and memory cost. The system is verified with TSMC 90 nm technology and implemented on a FPGA system which can achieve output resolution of 1920x1080 full HD resolution at 60 frames per second. The proposed system is working at a clock frequency of 148.36 MHz. The proposed architecture performs a dictionary learning-based algorithm that generates high-resolution images from low-resolution images using only numbers of line buffers. This real-time framebuffer-free system makes it possible to integrate super-resolution operation into image sensors or display drivers carrying out computational photography in display system.