重写屏幕找图算法,保证准确性

This commit is contained in:
fofolee 2025-02-17 11:46:08 +08:00
parent 2476037c09
commit 174d3ed7e7
2 changed files with 89 additions and 204 deletions

View File

@ -1,68 +1,10 @@
const { nativeImage } = require("electron");
const { captureScreen } = require("./screenCapture");
// 将颜色值映射到8个区间
function mapColorValue(val) {
if (val > 223) return 7; // [224 ~ 255]
if (val > 191) return 6; // [192 ~ 223]
if (val > 159) return 5; // [160 ~ 191]
if (val > 127) return 4; // [128 ~ 159]
if (val > 95) return 3; // [96 ~ 127]
if (val > 63) return 2; // [64 ~ 95]
if (val > 31) return 1; // [32 ~ 63]
return 0; // [0 ~ 31]
}
// 计算图像特征向量
function calculateFeatureVector(
buffer,
width,
height,
startX = 0,
startY = 0,
w = width,
h = height
) {
// 8^4 = 4096 维向量表示RGBA各8个区间的组合
const vector = new Array(8 * 8 * 8 * 8).fill(0);
for (let y = startY; y < startY + h; y++) {
for (let x = startX; x < startX + w; x++) {
const idx = (y * width + x) * 4;
// 计算四个通道的量化值
const r = mapColorValue(buffer[idx]);
const g = mapColorValue(buffer[idx + 1]);
const b = mapColorValue(buffer[idx + 2]);
const a = mapColorValue(buffer[idx + 3]);
// 计算在向量中的位置
const vectorIdx = r * 512 + g * 64 + b * 8 + a;
vector[vectorIdx]++;
}
}
return vector;
}
// 计算余弦相似度
function calculateCosineSimilarity(v1, v2) {
let dotProduct = 0;
let norm1 = 0;
let norm2 = 0;
for (let i = 0; i < v1.length; i++) {
dotProduct += v1[i] * v2[i];
norm1 += v1[i] * v1[i];
norm2 += v2[i] * v2[i];
}
return dotProduct / (Math.sqrt(norm1) * Math.sqrt(norm2));
}
// 获取显示器缩放比例
function getDisplayScale() {
// MacOS 上要考虑缩放比例
if (process.platform === "darwin") {
// 在 macOS 上,通过比较实际分辨率和报告的分辨率来计算缩放比例
const primaryDisplay = utools.getPrimaryDisplay();
const { scaleFactor } = primaryDisplay;
return scaleFactor;
@ -71,153 +13,94 @@ function getDisplayScale() {
}
// 在屏幕上查找图片
async function findImage(targetImageData, options = {}) {
try {
// 获取屏幕截图
const screenDataUrl = await captureScreen();
if (!screenDataUrl) return null;
async function findImage(subDataURL, options = {}) {
const mainDataURL = await captureScreen();
if (!mainDataURL) return null;
// 解析主图和子图
const mainImg = nativeImage.createFromDataURL(mainDataURL);
const subImg = nativeImage.createFromDataURL(subDataURL);
// 获取显示器缩放比例
const scale = getDisplayScale();
// 获取图像基本信息
const mainSize = mainImg.getSize();
const subSize = subImg.getSize();
// 读取屏幕截图
const screenImage = nativeImage.createFromDataURL(screenDataUrl);
const screenBuffer = screenImage.toBitmap();
const { width: actualWidth, height: actualHeight } = screenImage.getSize();
// 获取像素数据返回BufferRGBA格式
const mainPixels = mainImg.getBitmap();
const subPixels = subImg.getBitmap();
// 计算缩放后的实际尺寸
const screenWidth = Math.round(actualWidth / scale);
const screenHeight = Math.round(actualHeight / scale);
// 边界检查
if (subSize.width > mainSize.width || subSize.height > mainSize.height) {
throw new Error("要查找图片尺寸大于屏幕");
}
// 从 base64 字符串创建目标图片
const targetImage = nativeImage.createFromDataURL(targetImageData);
const targetBuffer = targetImage.toBitmap();
const { width: targetWidth, height: targetHeight } = targetImage.getSize();
// 预提取子图首像素值(优化点)
const firstSubPixel = [
subPixels[0], // R
subPixels[1], // G
subPixels[2], // B
subPixels[3], // A
];
// 计算目标图片的特征向量
const targetVector = calculateFeatureVector(
targetBuffer,
targetWidth,
targetHeight
);
// 主图遍历边界
const maxX = mainSize.width - subSize.width;
const maxY = mainSize.height - subSize.height;
// 设置匹配阈值
const threshold = options.threshold || 0.9;
// 遍历主图每个可能的位置
for (let y = 0; y <= maxY; y++) {
for (let x = 0; x <= maxX; x++) {
// 快速检查首像素(性能优化关键)
const mainOffset = (y * mainSize.width + x) * 4;
if (
mainPixels[mainOffset] !== firstSubPixel[0] ||
mainPixels[mainOffset + 1] !== firstSubPixel[1] ||
mainPixels[mainOffset + 2] !== firstSubPixel[2] ||
mainPixels[mainOffset + 3] !== firstSubPixel[3]
) {
continue;
}
let bestMatch = null;
let bestSimilarity = 0;
// 完整像素比对
let match = true;
for (let subY = 0; subY < subSize.height; subY++) {
for (let subX = 0; subX < subSize.width; subX++) {
// 计算像素位置
const mainPixelPos = ((y + subY) * mainSize.width + (x + subX)) * 4;
const subPixelPos = (subY * subSize.width + subX) * 4;
// 使用滑动窗口搜索
const stepSize = Math.round(8 * scale); // 根据缩放比例调整步长
for (let y = 0; y <= actualHeight - targetHeight; y += stepSize) {
for (let x = 0; x <= actualWidth - targetWidth; x += stepSize) {
// 计算当前区域的特征向量
const regionVector = calculateFeatureVector(
screenBuffer,
actualWidth,
actualHeight,
x,
y,
targetWidth,
targetHeight
);
// 计算相似度
const similarity = calculateCosineSimilarity(
targetVector,
regionVector
);
// 更新最佳匹配
if (similarity > bestSimilarity) {
bestSimilarity = similarity;
bestMatch = { x: Math.round(x / scale), y: Math.round(y / scale) };
// 如果相似度已经很高,进行精确搜索
if (similarity >= threshold) {
// 在周围进行精确搜索,注意搜索范围也要考虑缩放
const searchRange = Math.round(4 * scale);
for (let dy = -searchRange; dy <= searchRange; dy++) {
for (let dx = -searchRange; dx <= searchRange; dx++) {
const newX = x + dx;
const newY = y + dy;
if (
newX < 0 ||
newY < 0 ||
newX > actualWidth - targetWidth ||
newY > actualHeight - targetHeight
) {
continue;
}
const preciseVector = calculateFeatureVector(
screenBuffer,
actualWidth,
actualHeight,
newX,
newY,
targetWidth,
targetHeight
);
const preciseSimilarity = calculateCosineSimilarity(
targetVector,
preciseVector
);
if (preciseSimilarity > bestSimilarity) {
bestSimilarity = preciseSimilarity;
bestMatch = {
x: Math.round(newX / scale),
y: Math.round(newY / scale),
};
}
}
}
// 精确匹配每个通道
if (
mainPixels[mainPixelPos] !== subPixels[subPixelPos] ||
mainPixels[mainPixelPos + 1] !== subPixels[subPixelPos + 1] ||
mainPixels[mainPixelPos + 2] !== subPixels[subPixelPos + 2] ||
mainPixels[mainPixelPos + 3] !== subPixels[subPixelPos + 3]
) {
match = false;
break;
}
}
if (!match) break;
}
// 如果找到足够好的匹配,提前返回
if (bestSimilarity >= threshold) {
const position = {
x: bestMatch.x,
y: bestMatch.y,
width: Math.round(targetWidth / scale),
height: Math.round(targetHeight / scale),
confidence: bestSimilarity,
};
clickImage(position, options.mouseAction);
return position;
}
if (match) {
const displayScale = getDisplayScale();
const position = {
x: Math.round(x / displayScale),
y: Math.round(y / displayScale),
width: Math.round(subSize.width / displayScale),
height: Math.round(subSize.height / displayScale),
};
clickImage(position, options.mouseAction);
return position;
}
}
// 如果没有找到足够好的匹配,但有最佳匹配且相似度不太低,也返回
if (bestMatch && bestSimilarity > threshold * 0.8) {
const position = {
x: bestMatch.x,
y: bestMatch.y,
width: Math.round(targetWidth / scale),
height: Math.round(targetHeight / scale),
confidence: bestSimilarity,
};
clickImage(position, options.mouseAction);
return position;
}
return null;
} catch (error) {
console.error("查找图片失败:", error);
return null;
}
return null;
}
const clickImage = (position, mouseAction) => {
// 计算中心点
const centerX = position.x + position.width / 2;
const centerY = position.y + position.height / 2;
const centerX = Math.round(position.x + position.width / 2);
const centerY = Math.round(position.y + position.height / 2);
// 根据配置执行鼠标动作
switch (mouseAction) {

View File

@ -36,7 +36,7 @@
<!-- 配置区域 -->
<div class="col-12 col-sm-4">
<div class="row q-col-gutter-sm">
<div class="row">
<!-- 从剪贴板读取按钮 -->
<div class="col-12">
<q-btn
@ -53,8 +53,13 @@
<!-- 匹配阈值设置 -->
<div class="col-12">
<NumberInput
v-model="argvs.threshold"
v-if="false"
:model-value="argvs.threshold"
@update:model-value="updateArgvs('threshold', $event)"
label="匹配阈值"
:min="0"
:max="1"
:step="0.1"
class="border-primary"
:command="{
icon: 'tune',
@ -64,20 +69,15 @@
<!-- 鼠标动作选择 -->
<div class="col-12">
<q-select
v-model="argvs.mouseAction"
<ButtonGroup
:is-collapse="false"
:model-value="argvs.mouseAction"
@update:model-value="updateArgvs('mouseAction', $event)"
:options="mouseActionOptions"
label="找到后"
class="border-primary"
dense
filled
emit-value
map-options
>
<template v-slot:prepend>
<q-icon name="mouse" />
</template>
</q-select>
</ButtonGroup>
</div>
</div>
</div>
@ -97,11 +97,13 @@
<script>
import { defineComponent } from "vue";
import NumberInput from "components/composer/common/NumberInput.vue";
import ButtonGroup from "components/composer/common/ButtonGroup.vue";
export default defineComponent({
name: "ImageSearchEditor",
components: {
NumberInput,
ButtonGroup,
},
props: {
@ -123,7 +125,7 @@ export default defineComponent({
],
defaultArgvs: {
imagePreview: "",
threshold: 0.9,
threshold: 1,
mouseAction: "none",
},
};
@ -266,8 +268,8 @@ export default defineComponent({
border: 2px dashed var(--q-primary);
border-radius: 8px;
display: flex;
max-height: 128px;
min-height: 128px;
max-height: 138px;
min-height: 138px;
justify-content: center;
align-items: center;
position: relative;