Upload 3 files

README.md

@@ -12,4 +12,4 @@ pinned: false
 
 # KAIRO.ai
 
-AI 기반 PNG 이미지 문서 오탈자 및 문법 교정 플랫폼 - EasyOCR + PaddleOCR 이중 적용
+AI 기반 PNG 이미지 교정 (OCR + GPT-4 Turbo + PPT 생성) 시스템입니다.

app_final_style.py

@@ -0,0 +1,135 @@
+
+import streamlit as st
+import os
+import zipfile
+import io
+from PIL import Image
+import easyocr
+from paddleocr import PaddleOCR
+import numpy as np
+import openai
+from pptx import Presentation
+from pptx.util import Inches, Pt
+from pptx.dml.color import RGBColor
+
+st.set_page_config(page_title="KAIRO.ai", layout="wide")
+
+# 헤더 타이틀
+st.markdown(
+    "<h1 style='text-align:center; font-size:4em;'>🧠 KAIRO.ai</h1>"
+    "<p style='text-align:center; color:gray;'>v0.9 | 마지막 업데이트: 2025-06-02</p>",
+    unsafe_allow_html=True
+)
+
+ocr_engines = st.multiselect("OCR 엔진 선택 (중복 가능)", ["EasyOCR", "PaddleOCR"], default=["EasyOCR"])
+checklist = st.multiselect("검수 항목 선택", ["오탈자", "띄어쓰기", "문법", "어휘"], default=["오탈자", "띄어쓰기"])
+uploaded_files = st.file_uploader("PNG 또는 ZIP 파일 업로드", type=["png", "zip"], accept_multiple_files=True)
+api_key = st.text_input("OpenAI API Key", type="password")
+
+def extract_png_from_zip(zip_file):
+    png_images = []
+    with zipfile.ZipFile(zip_file) as z:
+        for f in z.namelist():
+            if f.lower().endswith(".png"):
+                with z.open(f) as file:
+                    image = Image.open(file)
+                    png_images.append((f, image.copy()))
+    return png_images
+
+def highlight_diff(original, corrected):
+    # 강조 구문 단순 로직
+    original_words = original.split()
+    corrected_words = corrected.split()
+    highlighted_original = []
+    highlighted_corrected = []
+
+    for o, c in zip(original_words, corrected_words):
+        if o != c:
+            highlighted_original.append(f"[빨]{o}[/빨]")
+            highlighted_corrected.append(f"[파]{c}[/파]")
+        else:
+            highlighted_original.append(o)
+            highlighted_corrected.append(c)
+    return " ".join(highlighted_original), " ".join(highlighted_corrected)
+
+def add_slide(prs, title_text, image, original, corrected):
+    slide = prs.slides.add_slide(prs.slide_layouts[6])
+    left = Inches(0.5)
+    top = Inches(0.5)
+    width = Inches(2.5)
+    height = Inches(2)
+
+    # 이미지 삽입
+    img_path = "/tmp/tempimg.png"
+    image.save(img_path)
+    slide.shapes.add_picture(img_path, left, top, width=width)
+
+    # 제목
+    title_box = slide.shapes.add_textbox(Inches(3.2), Inches(0.5), Inches(6), Inches(0.5))
+    title_frame = title_box.text_frame
+    title_frame.text = title_text
+
+    # 표 삽입
+    rows, cols = 2, 2
+    table = slide.shapes.add_table(rows, cols, Inches(3.2), Inches(1.2), Inches(6), Inches(2)).table
+    table.columns[0].width = Inches(3)
+    table.columns[1].width = Inches(3)
+    table.cell(0,0).text = "인식된 내용"
+    table.cell(0,1).text = "교정된 내용"
+
+    # 강조 텍스트
+    o_text, c_text = highlight_diff(original, corrected)
+    table.cell(1,0).text = o_text.replace("[빨]", "").replace("[/빨]", "")
+    table.cell(1,1).text = c_text.replace("[파]", "").replace("[/파]", "")
+
+    # 색상 강조 적용
+    for i, cell in enumerate([table.cell(1,0), table.cell(1,1)]):
+        for p in cell.text_frame.paragraphs:
+            for r in p.runs:
+                if "[빨]" in o_text and i == 0:
+                    r.font.color.rgb = RGBColor(255, 0, 0)
+                if "[파]" in c_text and i == 1:
+                    r.font.color.rgb = RGBColor(0, 176, 240)
+
+if uploaded_files and api_key:
+    all_images = []
+    for file in uploaded_files:
+        if file.name.lower().endswith(".png"):
+            image = Image.open(file)
+            all_images.append((file.name, image))
+        elif file.name.lower().endswith(".zip"):
+            all_images += extract_png_from_zip(file)
+
+    st.success(f"{len(all_images)}개 이미지 분석 시작")
+
+    prs = Presentation()
+
+    for fname, image in all_images:
+        full_text = ""
+        if "EasyOCR" in ocr_engines:
+            reader = easyocr.Reader(['ko', 'en'])
+            result = reader.readtext(np.array(image), detail=0)
+            full_text += "
+".join(result)
+        if "PaddleOCR" in ocr_engines:
+            paddle = PaddleOCR(use_angle_cls=True, lang='korean')
+            result = paddle.ocr(np.array(image), cls=True)
+            paddle_text = ""
+            for line in result[0]:
+                paddle_text += line[1][0] + "
+"
+            full_text += "
+" + paddle_text
+
+        openai.api_key = api_key
+        prompt = f"다음 문장을 교정해줘. 기준: {', '.join(checklist)}\n\n{full_text}"
+        response = openai.ChatCompletion.create(
+            model="gpt-4-turbo",
+            messages=[{"role": "user", "content": prompt}]
+        )
+        corrected = response.choices[0].message.content
+        add_slide(prs, fname, image, full_text, corrected)
+
+    pptx_path = "/mnt/data/KAIRO_FINAL_STYLE_v1.pptx"
+    prs.save(pptx_path)
+    st.download_button("📥 교정 결과 다운로드", data=open(pptx_path, "rb").read(), file_name="KAIRO_FINAL_STYLE_v1.pptx")

requirements.txt

@@ -1,7 +1,6 @@
 
 streamlit
+pytesseract
 Pillow
-openai
 python-pptx
-easyocr
-paddleocr
+openai