高解析影像一直以來都是人們所追求的，由於從高解析影像中可以取得更多的資訊，例如為高解析衛星影像具備較佳的分類區域與分析。一般而言，解析度通常以增加感測器的密度來達成，然而設備與設計成本相當高。尤其衛星高密度感測器必須冒更大的風險。於是我們選擇以多張影像合併方式來發展有效的超解析演算法來達到解析度提升的目標。 先假設所取得的低解析衛星影像間的移動皆在相同平面上，將低解析影像轉換至頻域，進行轉動與移動估測，再將影像對應到高解析格點，再用雙立方內插重建高解析影像。因為衛星影像龐大會造成計算量大幅增加，於是我們用已知衛星參數與區域影像內容進行過濾來降低移動估測與內插像元計算量。結果顯示確實能在維持重建品質前提下降低運算量。 People always desire high resolution image. The reason is higher resolution image can be obtained more information. For example, high resolution satellite images include better classification to identify and analyze. Generally, resolution enhancement is usually completed by increasing density of sensors. However, the additional costs of equipment and design are quite high. Especially, high density satellite sensors must take a big risk. So we choose multiple images composing to develop efficient super-resolution method for achieving resolution enhancement. We use frequency model to realize super-resolution. Assumed motion of the low resolution satellite images are all on the same plane. Then, estimate rotation and shift in frequency domain. After estimation, we compensate motion and stick on the high resolution grid. Bicubic interpolation method is used to reconstruct high resolution images. Because of the computation cost, we develop a satellite image information parameters filtering to decrease the estimation and interpolation computation. The results show that our method can decrease computation and keep the reconstruction quality.
