本論文中，我們提出了一種改良式的粒子群演算法，名為增減式慣性權重粒子群演算法(particle swarm optimization with increasing - decreasing inertia weights, IDWPSO)，並應用IDWPSO於影像壓縮。標準粒子群演算法中每一個體使用共同的慣性權重，而本文所提的方法，使粒子能自適應性地產生自己的慣性權重，在粒子群最佳化初期，透過遞增慣性權重，可以更有效的從局部探索開始，逐漸演化，等到集中收斂至一階段後，再將其它轉換成二階遞減模式以求迅速的把其它個體帶往全域最佳解。接著我們處理影像壓縮的問題，利用IDWPSO方法，獲取更好的壓縮率。由於我們得知最小平方法產生的預測誤差，往往出現在影像邊界相交處，所以我們在偵測到影像邊界時，採取IDWPSO預測器來提升預測的精確性，以防止耗費大量的運算，減少系統的運算的複雜度。從實驗結果證實，所提出的IDWPSO可以大幅的增進預測的精確性，最後在位元率(bit/pixel)的比較方面，與MED (Median Edge Detector, MED)相比改善了約7%、與GAP (Gradient-Adjusted Prediction, GAP)相比改進了約4%，也比EDP (Edge-directed Prediction, EDP)相比降低了約2%，證實所提出的演算法的確能有效提高影像編碼的效能。 In this thesis, we propose a modified optimization algorithm which is called particle swarm optimization with increasing-decreasing inertia weights (IDWPSO). Unlike the standard PSO algorithm, the proposed IDWPSO utilizes different weights for different particles. Initially, a small inertia weight is used for each particle to begin a global search. Then the individual inertia weights are respectively increasing linearly for more effective local searches. Finally, the inertia weights are switched to a larger value and then decreased quadratically to find a convergent optimum. Afterwards, the IDWPSO is applied to image coding problem as an image predictor. The IDWPSO predictor will be operated only when an edge is detected. The experimental results show that the proposed lossless image coding approach obtains more accurate image prediction. And better bit-rate compression is also obtained. As seen in the experiments, the IDWPSO is a 7% improvement over the MED (Median Edge Detector, MED), 4% over GAP (Gradient-Adjusted Prediction, GAP), and 2% over EDP (Edge-directed Prediction, EDP). These demonstrate the effectiveness of the proposed IDWPSO for the image coding.
