計畫名稱 : 塗料品質之彩包數位映像處理
摘要:
由計畫主持人和他過去博士班的學生,在早先所發展建立的「影像辨識方法」,針對鏽蝕程度評估,大致可分為兩種:NFRA (Neuro-Fuzzy Recognition Approach,模糊類神經網路辨識方法)和 SKMA(Simplified K-Means algorithm,簡易 K 平均演算法)。 NFRA 的方法是採用人工智慧技術,將鏽蝕的圖像由背景去除;而 SKMA 則是採用「K 平 均演算法」,透過數位化影像將目標圖像與背景圖像分離。即使這兩種方法有不同的製 程,但有一共通處:皆是將原來彩色的影像轉換為灰階,進而處理灰階影像,而非直接 處理彩色影像。然而這些方法在擷取數位化影像時,卻有部分限制,包括有:非均勻照度、低對比 數位影像、及畫面表層干擾的因素。因此,此計畫最主要的目標,是將色域理論應用於數位化彩色影像的處理過程,以 及發展一個控制表層塗漆品質的自動化系統。鏽蝕程度評估系統將透過以下七個步驟來完成:
摘要:
由計畫主持人和他過去博士班的學生,在早先所發展建立的「影像辨識方法」,針對鏽蝕程度評估,大致可分為兩種:NFRA (Neuro-Fuzzy Recognition Approach,模糊類神經網路辨識方法)和 SKMA(Simplified K-Means algorithm,簡易 K 平均演算法)。 NFRA 的方法是採用人工智慧技術,將鏽蝕的圖像由背景去除;而 SKMA 則是採用「K 平 均演算法」,透過數位化影像將目標圖像與背景圖像分離。即使這兩種方法有不同的製 程,但有一共通處:皆是將原來彩色的影像轉換為灰階,進而處理灰階影像,而非直接 處理彩色影像。然而這些方法在擷取數位化影像時,卻有部分限制,包括有:非均勻照度、低對比 數位影像、及畫面表層干擾的因素。因此,此計畫最主要的目標,是將色域理論應用於數位化彩色影像的處理過程,以 及發展一個控制表層塗漆品質的自動化系統。鏽蝕程度評估系統將透過以下七個步驟來完成:
- 鋼樑橋表層塗漆影像
- 色域最佳化
- 系統調整一致化
- 產生長條圖
- 目標區域的劃分
- 影像重建
- 鏽蝕程度評估 這個欲發展的系統,其驗證會確保質與量的並重。量的部份採用數理方法來推算鏽蝕的百分比;而質的部份則透過視覺化判定,及數位處理後之結果,判定是否能可靠地反應彩色表層塗漆的影像。
Project : Automated Digital Color Image Processing for Coating Quality
Abstrac:
Previously developed image recognition methods by the Principal Investigator (PI) and his former Ph.D. students for rust defect assessment can be summarized as two: the NFRA (Neuro-Fuzzy Recognition Approach) method and the SKMA (Simplified K-Means algorithm) method. The NFRA method uses artificial intelligence techniques to separate rust pixels from background pixels. The SKMA method segments object pixels and background pixels in a digitized image using a statistical method, called the K-means algorithm. Even if both methods pass through different processing procedures, one common thing is that they first convert original color images to grayscale images and further process the grayscale images. They do not process color image directly.
The system are often encountered difficulties while acquiring digital images under environmental conditions such as non-uniform illuminations, low-contrast digital images, and noises on painting surfaces. The purpose of this proposal is to explore the use of color space theories for processing digital color image and developing an automated system on controlling coating quality.
The rust defect assessment system will be realized by passing through seven steps: 1) steel bridge coating image, 2) optimal color space, 3) coordinate system adjustment, 4) histogram generation, 5) separation of target areas, 6) image reconstruction, and 7) assessment of defects.
Abstrac:
Previously developed image recognition methods by the Principal Investigator (PI) and his former Ph.D. students for rust defect assessment can be summarized as two: the NFRA (Neuro-Fuzzy Recognition Approach) method and the SKMA (Simplified K-Means algorithm) method. The NFRA method uses artificial intelligence techniques to separate rust pixels from background pixels. The SKMA method segments object pixels and background pixels in a digitized image using a statistical method, called the K-means algorithm. Even if both methods pass through different processing procedures, one common thing is that they first convert original color images to grayscale images and further process the grayscale images. They do not process color image directly.
The system are often encountered difficulties while acquiring digital images under environmental conditions such as non-uniform illuminations, low-contrast digital images, and noises on painting surfaces. The purpose of this proposal is to explore the use of color space theories for processing digital color image and developing an automated system on controlling coating quality.
The rust defect assessment system will be realized by passing through seven steps: 1) steel bridge coating image, 2) optimal color space, 3) coordinate system adjustment, 4) histogram generation, 5) separation of target areas, 6) image reconstruction, and 7) assessment of defects.