CameraMachine/python/service/shoe_main_image_compose.py

# -*- coding: utf-8 -*-
"""
鞋类主图合成

基于底图模板（如“粉色布景 + 左侧浅色弧形区域”的底图，即需求中的图2），合成三块区域：

    A区域：顶部 logo —— 固定顶部位置，宽度不超过底图浅色区域在该位置的宽度
    B区域：鞋款色卡列 —— 不同款色的鞋子（数量不固定：1款、5款、更多均可），
           在垂直方向按数量等分放置，每只鞋下方（或右侧）放置该款鞋的颜色文本
    C区域：最底部文本信息 —— 圆角信息条（货号/尺码/装箱等）

浅色区域宽度由本模块自动检测（对模板左侧浅色像素逐行求右缘），无需手工量坐标。
仅依赖 Pillow + numpy。logo、鞋子图建议使用带透明通道的 PNG（合成前自动裁切到 alpha 最小边）。

用法示例：
    from shoe_main_image_compose import compose_shoe_main_image

    compose_shoe_main_image(
        template=r"底图.png",
        logo=r"logo.png",
        shoes=[
            {"image": r"鞋_米白.png", "color": "米白"},
            {"image": r"鞋_粉色.png", "color": "粉色"},
            {"image": r"鞋_蓝色.png", "color": "蓝色"},
        ],
        info_lines=[
            ("货号(中童)", "1025B-17"),
            ("尺码 Size", "24-29#"),
            ("装箱 Box", "60双"),
        ],
        output_path=r"output.png",
    )
"""
import os, settings, requests
from io import BytesIO

import numpy as np
from PIL import Image, ImageDraw, ImageFont


# ---------------------------------------------------------------- 基础工具 --

def _to_image(im_or_path):
    """path 或 PIL.Image -> PIL.Image"""
    if isinstance(im_or_path, str):
        return Image.open(im_or_path)
    return im_or_path


def _load_font(font_path, size):
    """加载中文字体，未指定时按 项目resources -> 系统字体 顺序回退"""
    candidates = []
    if font_path:
        candidates.append(font_path)
    here = os.path.dirname(os.path.abspath(__file__))
    candidates += [
        os.path.join(here, "..", "resources", "ttf", "simhei.ttf"),
        os.path.join("resources", "ttf", "simhei.ttf"),
        r"C:\Windows\Fonts\simhei.ttf",
        r"C:\Windows\Fonts\msyh.ttc",
    ]
    for p in candidates:
        if p and os.path.exists(p):
            try:
                return ImageFont.truetype(p, size)
            except Exception:
                continue
    return ImageFont.load_default()


def _save(canvas, path):
    d = os.path.dirname(os.path.abspath(path))
    if d:
        os.makedirs(d, exist_ok=True)
    ext = os.path.splitext(path)[1].lower()
    if ext in (".jpg", ".jpeg"):
        canvas.convert("RGB").save(path, format="JPEG", quality=92)
    else:
        canvas.save(path, format="PNG")  # 保留 RGBA 透明通道


def _detect_band_widths(canvas, band_color=None, tolerance=30):
    """
    检测底图左侧浅色区域（如奶油色弧带）的逐行右缘。

    返回 (widths, band_color)：
        widths  np.array, len=H, widths[y] = 第 y 行浅色区域的右缘 x
        band_color 实际使用的浅色基准色 (r, g, b)

    band_color 为 None 时，从模板左上角采样取中值作为基准色。
    """
    a = np.asarray(canvas)
    rgb = a[:, :, :3].astype(float)                 # 只用 RGB 通道做颜色判定
    alpha = a[:, :, 3] if a.shape[2] >= 4 else None  # 有透明通道时排除透明像素
    H, W = rgb.shape[:2]
    if band_color is None:
        sample = rgb[4:64, 4:44, :].reshape(-1, 3)
        bc = np.median(sample, axis=0)
    else:
        bc = np.asarray(band_color, dtype=float)

    # 双条件匹配：颜色距离 + 色度差（通道极差 max-min）。
    # 单靠颜色距离时，比奶油色"更白"的纯白背景（距离 < tolerance）会被误判成浅色区，
    # 导致带宽被高估到 W*0.6 上限。奶油色带暖调（极差≈24），纯白≈0、深藏青≈15，
    # 用 |像素极差 - 基准极差| < cf_tol 即可排除白底，同时兼容 奶油→藏青 等其它配色。
    dist = np.abs(rgb - bc).max(axis=2)
    band_cf = float(bc.max() - bc.min())
    cf = rgb.max(axis=2) - rgb.min(axis=2)
    cf_tol = max(6.0, band_cf * 0.5)
    match = (dist < tolerance) & (np.abs(cf - band_cf) < cf_tol)
    if alpha is not None:
        match = match & (alpha > 128)  # 透明区（模板右侧）不算浅色区
    match = match.astype(np.int8)

    win = 21
    kernel = np.ones(win)
    limit = int(W * 0.6)  # 浅色区域不会超过图宽的 60%
    widths = np.zeros(H, dtype=int)
    for y in range(H):
        c = np.convolve(match[y], kernel, mode="same")
        cands = np.where(c >= win * 0.6)[0]
        cands = cands[cands < limit]
        widths[y] = int(cands.max()) if cands.size else 0

    # 中值平滑，消除单行噪声
    sm = np.zeros_like(widths)
    for y in range(H):
        lo, hi = max(0, y - 6), min(H, y + 7)
        sm[y] = int(np.median(widths[lo:hi]))
    widths = sm

    # 兜底：没检测到浅色区域时按图宽 32% 处理
    if int(widths.max()) < 40:
        widths = np.full(H, int(W * 0.32), dtype=int)
    return widths, tuple(int(v) for v in bc)


def _dominant_color(im):
    """取带透明通道图片的主色（用于颜色文本前的圆点）"""
    try:
        a = np.asarray(im.convert("RGBA"))
        px = a[a[:, :, 3] > 128][:, :3].astype(np.int32)
        if px.size == 0:
            return (120, 120, 120)
        q = px // 24 * 24
        key = q[:, 0] * 65536 + q[:, 1] * 256 + q[:, 2]
        counts = np.bincount(key)
        order = np.argsort(-counts)[:5]
        sel = np.isin(key, np.unique(key)[order])
        col = q[sel].mean(axis=0)
        return tuple(int(v) for v in col)
    except Exception:
        return (120, 120, 120)


def _draw_color_label(draw, font, text, x, y, text_color, dot=None, align="left"):
    """
    绘制 “● 颜色名” 标签。
    x, y 为文本视觉左上角；align="center" 时 x 为整体（圆点+文本）的水平中心。
    dot 非 None（颜色元组）时在文本前画小圆点。
    """
    tb = draw.textbbox((0, 0), text, font=font)
    tw, th = tb[2] - tb[0], tb[3] - tb[1]
    if dot is not None:
        dot_d = max(6, int(th * 0.55))
        unit_w = dot_d + 4 + tw
        if align == "center":
            x = int(x - unit_w / 2)
        cy = y + th // 2
        draw.ellipse([x, cy - dot_d // 2, x + dot_d, cy + dot_d // 2], fill=tuple(dot))
        x += dot_d + 4
    draw.text((x - tb[0], y - tb[1]), text, font=font, fill=tuple(text_color))


def _paste_alpha(canvas, im, x, y):
    """按 alpha 通道粘贴"""
    im = im.convert("RGBA")
    canvas.paste(im, (int(x), int(y)), im.getchannel("A"))


def _trim_alpha(im, threshold=8):
    """裁切透明底图像的最小边（alpha>threshold 的包围盒），无透明通道则原样返回"""
    if im.mode != "RGBA":
        return im
    a = np.asarray(im)[:, :, 3]
    ys, xs = np.where(a > threshold)
    if xs.size == 0:
        return im
    return im.crop((int(xs.min()), int(ys.min()), int(xs.max()) + 1, int(ys.max()) + 1))


def upload_image_by_io(image_pil: Image) -> str:
    # post_headers = {"Authorization": settings.Authorization}
    im = image_pil
    # 带透明通道（RGBA）时存 PNG 以保留透明；否则存 JPEG
    has_alpha = im.mode in ("RGBA", "LA", "PA")
    fmt, ext, ctype = ("PNG", ".png", "image/png") if has_alpha else ("JPEG", ".jpg", "image/jpeg")
    img = BytesIO()
    try:
        im.save(img, format=fmt)  # format: PNG or JPEG
    except Exception:
        img = BytesIO()
        im.save(img, format="PNG")  # 兜底：RGBA 无法存 JPEG 时退化为 PNG
        ext, ctype = ".png", "image/png"
    img.seek(0)  # rewind to the start
    try:
        url = settings.DOMAIN + "/api/upload"
        resultData = requests.post(
            url, files={"file": ("image" + ext, img, ctype)},
            timeout=100
        ).json()
        return resultData["data"]["url"]
    except Exception as e:
        print("upload_image_by_io error:", e)
        return None


# ---------------------------------------------------------------- 主函数 --

def compose_shoe_main_image(
        template,
        logo=None,
        data=None,
        shoes=None,
        info_lines=None,
        output_path=None,
        upload=False,  # True=合成后上传到 OSS（upload_image_by_io）并返回链接地址
        font_path=None,
        # ---- 浅色区域检测 ----
        band_color=None,  # None=自动从模板左上角检测；可显式传 (r, g, b)
        band_tolerance=30,  # 浅色判定容差（最大通道差）
        # ---- A区域：logo ----
        logo_top=8,  # 顶部固定位置 y
        logo_left=8,  # 左侧位置 x
        logo_width_ratio=0.95,  # logo 最大宽度 = 浅色区域宽度(该位置) * 比例
        logo_width=None,  # 指定目标宽度（仍受浅色区域宽度约束）
        # ---- B区域：鞋款色卡列 ----
        shoes_top=None,  # None=logo 下方
        shoes_bottom=None,  # None=信息条上方
        cell_pad=4,  # 每格上下留白
        shoe_width_limit_ratio=1.15,  # 鞋子最大宽度 = 该格浅色区域宽度 * 比例（可略溢出）
        shoe_scale=1.0,  # 鞋子最终尺寸缩放系数：1.0=自动适配大小，>1 放大，<1 缩小
        label_pos="bottom",  # 颜色文本位置："bottom"下方 | "right"右侧（图1样式）
        label_font_size=15,
        label_color=(60, 60, 60),
        label_dot=True,  # 颜色文本前是否画小圆点（取鞋子主色）
        label_gap=5,  # 鞋与颜色文本的间距
        shoe_center_offset=0,  # 鞋子中心相对浅色区域中心的水平偏移
        # ---- C区域：底部文本信息 ----
        info_bar=True,  # True=画圆角信息条；False=只画文字
        info_bar_color=(112, 175, 217),
        info_text_color=(255, 255, 255),  # 条内文字色
        info_plain_text_color=(80, 80, 80),  # info_bar=False 时的文字色
        info_bar_height=None,  # 信息条总高度；None=由 info_bar_pad_y 决定（文本高 + 上下间距*2）
        info_bar_pad_y=5,  # 信息条内文字的上下间距（上下内边距，可调节）
        info_bar_gap=7,
        info_bar_width=None,  # None=每条按自身文本自适应宽度
        info_left=0,
        info_bottom_margin=16,
        info_font_size=14,
        info_pad_x=12,
        # ---- 调试 ----
        debug=False,  # 在输出旁另存 *_debug.png（浅色右缘/各区域框）
):
    """
    鞋类主图合成。

    参数：
        template    底图（图2）路径或 PIL.Image
        logo        A区域 logo 路径或 PIL.Image（建议透明 PNG），None=不放
        data        合并入参（推荐）：一个字典，同时包含 logo、底部信息和鞋款列表，如
                    {"logo": "logo.png",
                     "number_mid": {"name": "货号(中童)", "value": "1025B-17"},
                     "size_mid":   {"name": "尺码 Size", "value": "24-29#"},
                     "number_big": {"name": "货号(大童)", "value": "1025A-17"},
                     "size_big":   {"name": "尺码 Size", "value": "30-35#"},
                     "box":        {"name": "装箱 Box", "value": "60双"},
                     "images": [{"image_path": "鞋1.png", "color": "米白"}, ...]}
                    logo 取自 data["logo"]（未直接传 logo 时）；B区域取 data["images"] 每个
                    元素的 image_path/color；C区域按 number_mid/size_mid/number_big/size_big/box
                    顺序取字段，缺省字段自动跳过。
                    logo 与 image_path 均支持 路径 或 PIL.Image；传 PIL 时直接使用该对象
                    （内部先转 RGBA 再裁切/缩放，不会改动传入的对象本身）。
        shoes       B区域鞋款列表（data 未提供时的独立入参），元素为
                    {"image": 路径/PIL.Image, "color": "米白"}；数量不固定
        info_lines  C区域文本（data 未提供时的独立入参），元素为 (标签, 值) 或纯字符串，如
                    [("货号(中童)", "1025B-17"), ("装箱 Box", "60双")]
        output_path 输出路径（.jpg/.png），None=不保存
        upload      True=合成后调用 upload_image_by_io 上传到 OSS，并返回链接地址
        其余参数见函数签名默认值。

    返回：upload=True 时为上传后的链接地址（str，上传失败为 None）；
          否则为合成结果 PIL.Image (RGB)
    """
    shoes = shoes or []
    info_lines = info_lines or []

    # ---------- 合并入参 data -> logo + shoes + info_lines ----------
    # data: {logo, number_mid, size_mid, number_big, size_big, box, images:[{image_path,color},...]}
    if data:
        if logo is None:
            logo = data.get("logo")
        if not shoes:
            shoes = [{"image": im.get("image_path"), "color": im.get("color", "")}
                     for im in data.get("images", [])]
        if not info_lines:
            for key in ("number_mid", "size_mid", "number_big", "size_big", "box"):
                fv = data.get(key)
                if fv:
                    info_lines.append((str(fv.get("name", "")), str(fv.get("value", ""))))

    # ---------- 底图（保留透明通道：模板右侧为透明区，输出需为 RGBA） ----------
    canvas = _to_image(template).convert("RGBA")
    W, H = canvas.size
    draw = ImageDraw.Draw(canvas)

    # ---------- 浅色区域检测 ----------
    band_width, _band_c = _detect_band_widths(canvas, band_color=band_color,
                                              tolerance=band_tolerance)

    debug_boxes = []

    # ---------- A区域：logo（宽度受浅色区域约束） ----------
    def _band_min(y0, y1):
        y0, y1 = max(0, int(y0)), min(H, int(y1))
        if y1 <= y0:
            y1 = y0 + 1
        return float(band_width[y0:y1].min())

    logo_bottom = int(logo_top) + 10
    if logo is not None:
        logo_im = _trim_alpha(_to_image(logo).convert("RGBA"))
        lw0, lh0 = logo_im.size
        # 第一遍：按预估高度求该位置浅色区域最小宽度
        target_w = min(logo_width if logo_width else 1e18,
                       _band_min(logo_top, logo_top + max(20, lh0)) * logo_width_ratio)
        scale = target_w / lw0
        if logo_width or scale < 1:
            logo_im = logo_im.resize((max(1, int(lw0 * scale)), max(1, int(lh0 * scale))),
                                     Image.LANCZOS)
        # 第二遍：按实际高度复核，防止 logo 过高时压过浅色区域
        lw1, lh1 = logo_im.size
        limit2 = _band_min(logo_top, logo_top + lh1) * logo_width_ratio
        if lw1 > limit2 and limit2 > 4:
            s2 = limit2 / lw1
            logo_im = logo_im.resize((max(1, int(lw1 * s2)), max(1, int(lh1 * s2))),
                                     Image.LANCZOS)
            lw1, lh1 = logo_im.size
        _paste_alpha(canvas, logo_im, logo_left, logo_top)
        logo_bottom = int(logo_top) + lh1
        debug_boxes.append((logo_left, logo_top, logo_left + lw1, logo_bottom, (255, 0, 255)))

    # ---------- C区域：底部信息条（先算位置，决定 B 区域下界） ----------
    info_top = H
    if info_lines:
        rows = []
        for line in info_lines:
            if isinstance(line, (tuple, list)):
                lab = str(line[0]) if len(line) > 0 else ""
                val = str(line[1]) if len(line) > 1 else ""
            else:
                lab, val = str(line), ""
            rows.append((lab, val))

        info_font = _load_font(font_path, info_font_size)

        def _tw(t):
            if not t:
                return 0
            b = draw.textbbox((0, 0), t, font=info_font)
            return b[2] - b[0]

        bb = draw.textbbox((0, 0), "国A", font=info_font)
        th = bb[3] - bb[1]
        # 信息条高度：显式 info_bar_height 优先；否则 = 文本高 + 上下间距*2
        bar_h = int(info_bar_height) if info_bar_height else th + 2 * int(info_bar_pad_y)
        row_widths = []
        for lab, val in rows:
            w = _tw(lab) + (8 + _tw(val) if val else 0)
            row_widths.append(w + 2 * int(info_pad_x))
        if info_bar_width:
            row_widths = [int(info_bar_width)] * len(rows)

        block_h = len(rows) * bar_h + (len(rows) - 1) * int(info_bar_gap)
        info_top = H - int(info_bottom_margin) - block_h

        y = info_top
        for (lab, val), bw_row in zip(rows, row_widths):
            if info_bar:
                # 左侧直角、右侧圆角：左段矩形 + 右端半圆
                x0, x1 = int(info_left), int(info_left + bw_row)
                r = min(bar_h // 2, (x1 - x0) // 2)
                draw.rectangle([x0, y, x1 - r, y + bar_h], fill=tuple(info_bar_color))
                draw.ellipse([x1 - 2 * r, y, x1, y + bar_h], fill=tuple(info_bar_color))
            tc = tuple(info_text_color) if info_bar else tuple(info_plain_text_color)
            ty = y + (bar_h - th) // 2 - bb[1]
            if lab:
                draw.text((int(info_left + info_pad_x), ty), lab,
                          font=info_font, fill=tc)
            if val:
                vx = int(info_left + bw_row - info_pad_x - _tw(val))
                draw.text((vx, ty), val, font=info_font, fill=tc)
            debug_boxes.append((info_left, y, info_left + bw_row, y + bar_h, (0, 128, 255)))
            y += bar_h + int(info_bar_gap)

    # ---------- B区域：鞋款色卡列（按数量垂直等分） ----------
    n = len(shoes)
    if n:
        area_top = int(shoes_top) if shoes_top is not None else logo_bottom + 12
        area_bottom = (int(shoes_bottom) if shoes_bottom is not None
                       else (info_top - 12 if info_lines else H - 10))
        if area_bottom - area_top < 30:
            raise ValueError(
                "B区域可用高度不足(area_top={}, area_bottom={})，请调整 shoes_top/shoes_bottom 或减少鞋款数量"
                .format(area_top, area_bottom))
        cell_h = (area_bottom - area_top) / n
        lab_font = _load_font(font_path, label_font_size)

        for i, item in enumerate(shoes):
            if isinstance(item, str):
                shoe_src, color_name = item, ""
            else:
                shoe_src = item.get("image")
                color_name = str(item.get("color", ""))
            shoe_im = _trim_alpha(_to_image(shoe_src).convert("RGBA"))
            sw0, sh0 = shoe_im.size

            cell_y0 = area_top + i * cell_h
            cell_cy = int(min(cell_y0 + cell_h / 2, H - 1))
            bw = float(band_width[cell_cy])  # 该格中心处浅色区域宽度
            margin = max(3, int(bw * 0.03))  # 距浅色区左右边缘的安全留白

            # 颜色文本尺寸（圆点 + 4px + 文本，与 _draw_color_label 保持一致）
            if color_name:
                lb = draw.textbbox((0, 0), color_name, font=lab_font)
                label_h = lb[3] - lb[1]
                label_tw = lb[2] - lb[0]
                dot_d = max(6, int(label_h * 0.55)) if label_dot else 0
                label_w = dot_d + 4 + label_tw
            else:
                label_h = label_tw = label_w = 0

            # 该格内鞋子可占用的最大盒子：已扣除留白与颜色文本，
            # 保证"鞋 + 颜色名称"整体都落在浅色区内、绝不超出。
            if label_pos == "bottom":
                box_w = bw - 2 * margin
                box_h = cell_h - 2 * int(cell_pad) - (label_h + int(label_gap) if color_name else 0)
            else:  # right：鞋在左、文本紧跟其右，两者都要在带内
                box_w = bw - 2 * margin - (label_w + int(label_gap) if color_name else 0)
                box_h = cell_h - 2 * int(cell_pad)
            box_w = max(10.0, min(box_w, bw * float(shoe_width_limit_ratio)))
            fit_scale = min(box_w / sw0, box_h / sh0)
            scale = fit_scale * float(shoe_scale)
            sw, sh = max(2, int(sw0 * scale)), max(2, int(sh0 * scale))
            if sw > box_w or sh > box_h:  # shoe_scale>1 时仍限制在盒内
                sw, sh = max(2, int(sw0 * fit_scale)), max(2, int(sh0 * fit_scale))
            if (sw, sh) != (sw0, sh0):
                shoe_im = shoe_im.resize((sw, sh), Image.LANCZOS)

            if label_pos == "bottom":
                sx = int(max(margin, (bw - sw) / 2 + float(shoe_center_offset)))
                block_h = sh + (int(label_gap) + label_h if color_name else 0)
                sy = int(cell_y0 + max(0, (cell_h - block_h) / 2))
            else:  # right：鞋贴左缘，文本紧跟鞋右、且不超过浅色区右缘
                sx = int(margin)
                sy = int(cell_y0 + max(0, (cell_h - sh) / 2))
                label_x = int(sx + sw + int(label_gap))

            _paste_alpha(canvas, shoe_im, sx, sy)
            if debug:
                debug_boxes.append((sx, sy, sx + sw, sy + sh, (0, 255, 0)))

            if color_name:
                dot_color = _dominant_color(shoe_im) if label_dot else None
                if label_pos == "bottom":
                    ly = sy + sh + int(label_gap)
                    _draw_color_label(draw, lab_font, color_name, int(sx + sw / 2), ly,
                                      label_color, dot=dot_color, align="center")
                else:
                    ly = sy + (sh - label_h) // 2
                    _draw_color_label(draw, lab_font, color_name, label_x, ly,
                                      label_color, dot=dot_color, align="left")

    # ---------- 调试叠加图 ----------
    if debug and output_path:
        dbg = canvas.copy()
        d = ImageDraw.Draw(dbg)
        pts = [(int(band_width[y]), y) for y in range(0, H, 8)]
        d.line(pts, fill=(255, 0, 0), width=1)
        for x0, y0, x1, y1, c in debug_boxes:
            d.rectangle([int(x0), int(y0), int(x1), int(y1)], outline=c, width=2)
        base, ext = os.path.splitext(output_path)
        dbg.convert("RGB").save(base + "_debug" + (ext or ".png"))

    # ---------- 保存 / 上传 / 返回 ----------
    # if output_path:
    #     _save(canvas, output_path)
    if upload:
        return upload_image_by_io(canvas)
    return canvas