识别具有深色背景的文本和数字

Question

基础镜像

​

我的Python 3源代码：

from wand.image import Image
from wand.display import display

with Image(filename='edited.png') as img:
    with Image(img) as negate:
        negate.alpha_channel = False           
        negate.transform_colorspace('gray')
        negate.resize(width=negate.width*2, height=negate.height*2, filter='lanczos', blur=0.5)
        negate.negate()
        negate.contrast_stretch(black_point=0.00, white_point=0.8, channel=None)        
        negate.sharpen(radius=0, sigma=1)
        display(negate)

准备好的图像到tesseract

​

我们需要识别数字和一些文本。我的解决方案很好，但80%有效。如何做得更好，有什么选择吗？

Answer 1

我不确定为什么你的解决方案80%有效，但如果你使用pytesseract 4.1.1版，结果将如下所示：

Horsepower 1,675 HPA Pen Value at 100m §92/654/600 Nm Soft Terrain Resistance 2.20

Top Forward Speed 67 KM/H Pen Value at 500m 584/645/600 wm Medium Terrain Resista 130

Top Reverse Speed 20 KM/H Pen Value at 1000m 575/635/600 Nw Firm Terrain Resistance 1.10

Hull Rotation Speed 26 DEG/S Pen Value at 2000m 552/610/600 Nw

Weight to Horsepower F 26.56 Detectability Range: Sti 407.08 a

Fire Chance 10.00% Armor Damage 5§35/535/535 He Detectability Range: Mo 426.04 _
Module Damage 190/190/190 He

Accuracy at 100m 0.26" Blast Radius 0/0/0m View Range 515 M

Aiming Time 2.10 sec Turret Rotation Speed a0 oes =f

Max Elevation 20 0&6 = Ammo Max Speed 1,700/1,575/1,140 mys Accuracy During Rotatic 0.83 :

Max Depression 9 DEG AmmoMax Range  3,500/3,500/2,800 Signal Range 690

Ammo Capacity 50 Hit Points 4,000 He

Reload Time 7.90 sec Yaw Limit -180 180 oes

Rate of Fire 7.59 RPM Shell Type 1 APFSDS

Shots Per Clip 1 Shell Type 2 APFSDS

Intra-clip reload N/A Shell Type 3 HEAT

我检查了几个单词，它与原始输入图像相匹配。

我使用的是opencv而不是wand

代码：

import cv2
import pytesseract

# Load the img
img = cv2.imread("XX2ry.png")

# Cvt to gry
gry = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# OCR detection
txt = pytesseract.image_to_string(gry, config="--psm 6")
print(txt)

也许你应该检查一下page segmentation methods