Chart Vision — Complete Pipeline

Overview

Unified skill covering the entire chart image pipeline in one place:

1. Rendering — produce high-quality candlestick chart images from OHLCV data
2. Preprocessing — clean, denoise, and normalize raw chart screenshots
3. Candlestick Vision — detect single and multi-candle patterns from images
4. Chart Pattern Vision — detect H&S, triangles, wedges, double tops from images
5. Pattern Scanner — algorithmic classical pattern detection from price data
6. Trendline & S/R Vision — detect trendlines, channels, S/R zones from images
7. Annotation Overlay — draw all analysis results back onto the chart image

Pipeline Order

chart-vision-renderer       → render OHLCV to PNG
chart-image-preprocessor    → clean, denoise, extract ROI
chart-pattern-vision        → detect candle + chart patterns from image
trendline-sr-vision         → detect S/R and trendlines from image
chart-pattern-scanner       → detect patterns from price data (no image needed)
chart-annotation-overlay    → draw everything back onto chart  ← final step

Reference Files

Stack

• mplfinance — candlestick rendering
• matplotlib 3.10 — rendering engine
• OpenCV 4.13 — all CV operations (Hough, LSD, morphology, edge detection, drawing)
• scikit-image 0.26 — region props, probabilistic Hough, restoration
• scipy 1.17 — peak finding, signal processing, clustering
• Pillow 12.1 — text rendering, image post-processing
• numpy 2.4 — array operations
• sklearn — RandomForest, GradientBoosting, calibration, TimeSeriesSplit (AI signal aggregation & ML)
• statsmodels — ADF, Granger causality (statistical analysis)

Quick Usage Examples

Render a chart from OHLCV data

import mplfinance as mpf
import pandas as pd

df = pd.read_csv("ohlcv.csv", index_col="date", parse_dates=True)
mpf.plot(df, type="candle", style="charles", volume=True,
         mav=(20, 50), savefig="chart.png", figsize=(14, 8))

Preprocess a chart screenshot

import cv2
import numpy as np

img = cv2.imread("screenshot.png")
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Remove grid lines
denoised = cv2.fastNlMeansDenoising(gray, h=15)
# Enhance edges for pattern detection
edges = cv2.Canny(denoised, 50, 150)
# Extract ROI (crop to chart area)
contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
largest = max(contours, key=cv2.contourArea)
x, y, w, h = cv2.boundingRect(largest)
chart_roi = img[y:y+h, x:x+w]

Detect support/resistance from image

from scipy.signal import find_peaks

# Project pixel intensities horizontally to find price levels
projection = np.mean(gray, axis=1)
peaks, props = find_peaks(-projection, distance=20, prominence=10)
sr_levels = peaks  # pixel y-coordinates of S/R lines

# Draw detected levels
for level in sr_levels:
    cv2.line(img, (0, level), (img.shape[1], level), (0, 255, 0), 1)

Annotate chart with analysis

from PIL import Image, ImageDraw, ImageFont

img = Image.open("chart.png")
draw = ImageDraw.Draw(img)
font = ImageFont.truetype("arial.ttf", 14)

# Draw buy signal
draw.text((entry_x, entry_y - 20), "BUY", fill="green", font=font)
draw.rectangle([sl_x-2, sl_y-2, sl_x+2, sl_y+2], fill="red")
draw.text((sl_x + 5, sl_y), f"SL: {sl_price:.5f}", fill="red", font=font)
draw.text((tp_x + 5, tp_y), f"TP: {tp_price:.5f}", fill="green", font=font)
img.save("annotated_chart.png")