chart-vision

# 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 | File | Contents | |------|----------| | `references/rendering.md` | Chart renderer: mplfinance + matplotlib, indicators, dark/light themes | | `references/preprocessing.md` | Image preprocessing: denoise, ROI extract, grid removal, contrast, color analysis | | `references/pattern-vision.md` | CV candlestick detection + classical chart pattern detection from images | | `references/trendline-sr.md` | Hough transforms, LSD, horizontal projection, S/R clustering, channels | | `references/scanner-and-annotation.md` | Price-data pattern scanner (swing-based) + annotation overlay drawing engine | ## 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 ```python 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 ```python 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 ```python 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 ```python 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") ```