劉老師的跨域創想工坊

 

📌 教學目標

• 使用 OpenCV 內建的 EAST 模型 (frozen_east_text_detection.pb)

• 載入一張圖片並執行文字區塊偵測

• 以矩形框框出偵測到的文字位置

---

✅ 教學環境需求

元件	說明
Python 3.x	建議 3.8 以上
OpenCV	安裝 opencv-contrib-python
模型檔案	frozen_east_text_detection.pb

安裝必要套件

pip install opencv-python opencv-contrib-python numpy

---

📁 下載 EAST 模型檔案

你可以使用以下指令下載：

wget https://github.com/oyyd/frozen_east_text_detection.pb/raw/master/frozen_east_text_detection.pb

或手動從 GitHub - EAST pretrained model 下載。

---

🧪 Python 實作程式碼

儲存為 east_text_detect.py：

import cv2
import numpy as np

# 讀取模型
net = cv2.dnn.readNet("frozen_east_text_detection.pb")

# 載入影像
image = cv2.imread("test.webp")
orig = image.copy()
(H, W) = image.shape[:2]

# 設定輸入尺寸（必須是32的倍數）
newW, newH = (320, 320)
rW = W / float(newW)
rH = H / float(newH)
resized = cv2.resize(image, (newW, newH))
blob = cv2.dnn.blobFromImage(resized, 1.0, (newW, newH),
                             (123.68, 116.78, 103.94), swapRB=True, crop=False)

# 模型輸出層
outputLayers = ["feature_fusion/Conv_7/Sigmoid", "feature_fusion/concat_3"]
net.setInput(blob)
(scores, geometry) = net.forward(outputLayers)

# 解碼函數：從模型輸出取出 boxes
def decode(scores, geometry, confThreshold):
    (numRows, numCols) = scores.shape[2:4]
    boxes = []
    confidences = []

    for y in range(numRows):
        scoresData = scores[0, 0, y]
        x0 = geometry[0, 0, y]
        x1 = geometry[0, 1, y]
        x2 = geometry[0, 2, y]
        x3 = geometry[0, 3, y]
        anglesData = geometry[0, 4, y]
        for x in range(numCols):
            if scoresData[x] < confThreshold:
                continue

            offsetX, offsetY = x * 4.0, y * 4.0
            angle = anglesData[x]
            cos = np.cos(angle)
            sin = np.sin(angle)
            h = x0[x] + x2[x]
            w = x1[x] + x3[x]
            endX = int(offsetX + cos * x1[x] + sin * x2[x])
            endY = int(offsetY - sin * x1[x] + cos * x2[x])
            startX = int(endX - w)
            startY = int(endY - h)

            boxes.append([startX, startY, endX, endY])
            confidences.append(float(scoresData[x]))

    return boxes, confidences

boxes, confidences = decode(scores, geometry, confThreshold=0.5)

# 轉換為 [x, y, w, h] 格式以套用 NMSBoxes
rects = []
for (startX, startY, endX, endY) in boxes:
    rects.append([startX, startY, endX - startX, endY - startY])

# 非極大值抑制
indices = cv2.dnn.NMSBoxes(rects, confidences, score_threshold=0.5, nms_threshold=0.4)

# 繪製結果
if len(indices) > 0:
    for i in indices.flatten():
        (startX, startY, endX, endY) = boxes[i]
        startX = int(startX * rW)
        startY = int(startY * rH)
        endX = int(endX * rW)
        endY = int(endY * rH)
        cv2.rectangle(orig, (startX, startY), (endX, endY), (0, 255, 0), 2)

# 顯示結果
cv2.imshow("Text Detection", orig)
cv2.waitKey(0)
cv2.destroyAllWindows()

---

📷 測試用圖片

請放置一張含有明顯文字的圖片命名為 test.webp 與 east_text_detect.py 放在同一資料夾中。

---

✅ 教學重點回顧

步驟	重點
1️⃣ 下載 EAST .pb 模型	一定要是 TensorFlow frozen graph
2️⃣ 使用 cv2.dnn.readNet()	輸入模型與影像尺寸需對應（32 倍數）
3️⃣ 執行 forward() 並解析 geometry	取得文字位置與角度
4️⃣ 非極大值抑制處理重疊框	cv2.dnn.NMSBoxesRotated
5️⃣ 顯示結果	使用 OpenCV rectangle 框出文字

---