角點(diǎn)與目標(biāo)檢測(cè)(Corners and Object Detection)
已經(jīng)可以從圖像中提取基于形狀的特征薯蝎，如何使用這一組特征來(lái)檢測(cè)整個(gè)對(duì)象，以山峰圖像角點(diǎn)檢測(cè)舉例：
假設(shè)想要在其他圖像中檢測(cè)這座山的方法，單個(gè)角點(diǎn)不足以在任何其他圖像中識(shí)別這座山峰楷拳，但是狂芋，我們可以定義一組這座山峰的形狀特征，將它們組合后蹭一個(gè)數(shù)組或者向量森篷，然后使用這組特征來(lái)查UN構(gòu)造一個(gè)山峰檢測(cè)器。
以這個(gè)梯形及其相應(yīng)的梯度為例 這個(gè)圖像已經(jīng)檢測(cè)好了邊緣豺型，以梯形中心為基準(zhǔn) 觀察每個(gè)單元的梯度方向仲智，我們就可以將這個(gè)數(shù)據(jù)平面化 創(chuàng)建出一個(gè)一維數(shù)組，這就是特征向量姻氨；同樣的操作可應(yīng)用于其它形狀 如這個(gè)圓圈钓辆，取梯度計(jì)算網(wǎng)格每個(gè)單元的梯度方向。我們就得到了這兩種形狀的特征向量。
實(shí)時(shí)特征檢測(cè)-ORB（Oriented Fast and Rotated Brief）
-
首先在圖像中找到稱為關(guān)鍵點(diǎn)的特殊區(qū)域岩馍，可以將關(guān)鍵點(diǎn)視為圖像中的一個(gè)容易辨識(shí)小區(qū)域碉咆，例如角點(diǎn)。
以貓的圖像舉例：
- orb為每個(gè)關(guān)鍵點(diǎn)計(jì)算相應(yīng)的特征向量蛀恩，orb算法創(chuàng)建僅包含1和0的特征向量疫铜，因此也被稱為二元特征向量
ORB，不僅速度快双谆，不受噪聲照明壳咕、和圖像變換，如旋轉(zhuǎn)的影響
FAST 特征提取
ORB特征檢測(cè)第一步第一步是找到圖像中的關(guān)鍵點(diǎn)顽馋，這是由FAST算法完成的谓厘。
-
粗提取
從圖像選取一個(gè)像素P,判斷P是不是特征點(diǎn)的方法：以P為圓心，畫(huà)一個(gè)半徑為3pixel的圓寸谜，如果圓周上有連續(xù)n個(gè)像素點(diǎn)灰度值比p點(diǎn)大或者小竟稳，則認(rèn)為p為特征點(diǎn).一般n設(shè)置12或者8。FAST之所以可以這么快熊痴，因?yàn)樗紫葯z測(cè)的1他爸、5、9果善、13位置上的灰度值诊笤，如果p是特征點(diǎn)，那么這四個(gè)位置至少有3個(gè)或者3個(gè)以上的像素都大于或者小于P點(diǎn)的灰度值巾陕。如果不滿足讨跟，則直接排除該點(diǎn)。
FAST發(fā)現(xiàn)的關(guān)鍵點(diǎn)提供有關(guān)圖像中定義邊緣對(duì)象的信息鄙煤，但是這些關(guān)鍵點(diǎn)只給出了一個(gè)邊緣位置晾匠，并不包括任何有關(guān)強(qiáng)度變化方向的信息
Brief 特征描述
Brief將關(guān)鍵點(diǎn)轉(zhuǎn)化為二進(jìn)制特征向量。BRIEF stands for binary robust independent elementary features馆类。
簡(jiǎn)而言之混聊，每個(gè)關(guān)鍵點(diǎn)由具有的二進(jìn)制特征向量描述弹谁，一個(gè)128-512位的字符串乾巧，只包含1和0。每一位只能容納一位二進(jìn)制预愤。
Brief怎么為關(guān)鍵點(diǎn)生成二進(jìn)制描述：
- 首先BRIEF使用高斯內(nèi)核平滑給定圖像沟于，防止對(duì)高頻噪聲過(guò)于敏感。
- 然后植康，選取一個(gè)關(guān)鍵點(diǎn)旷太，在關(guān)鍵點(diǎn)附近定義的鄰域內(nèi)選擇一對(duì)隨機(jī)像素，關(guān)鍵點(diǎn)周圍的鄰域稱為補(bǔ)丁，它是具有一定像素寬度和高度的正方形供璧。
隨機(jī)對(duì)中的第一個(gè)像素存崖，這里顯示的是藍(lán)色正方形，是從以關(guān)鍵點(diǎn)為中心的高斯分布繪制的睡毒，具有標(biāo)準(zhǔn)偏差或Sigma的擴(kuò)展来惧。隨機(jī)對(duì)中的第二個(gè)像素，這里顯示為黃色正方形演顾，是從以該第一像素為中心的高斯分布繪制的供搀，標(biāo)準(zhǔn)偏差Sigma大于2。經(jīng)驗(yàn)表明钠至，這種高斯選擇提高了特征匹配率葛虐。
然后，BRIEF通過(guò)如下比較兩個(gè)像素的亮度開(kāi)始為關(guān)鍵點(diǎn)構(gòu)造二進(jìn)制描述符：如果第一像素比第二像素亮棉钧，則它將數(shù)值1分配給描述符中的對(duì)應(yīng)位屿脐，否則它分配的值為零。
在這個(gè)例子中宪卿，我們看到摄悯，第二個(gè)像素比第一個(gè)像素亮，所以我們給特征向量的第一個(gè)比特位賦值為0愧捕。特征向量的第一位對(duì)應(yīng)于該關(guān)鍵點(diǎn)的第一對(duì)隨機(jī)點(diǎn)奢驯。
現(xiàn)在，對(duì)于相同的關(guān)鍵點(diǎn)次绘，BRIEF選擇一個(gè)新的隨機(jī)像素對(duì)瘪阁，比較它們的亮度并將一或零分配給下一個(gè)比特和特征向量。在我們的例子中邮偎，我們看到現(xiàn)在第一個(gè)像素比第二個(gè)更亮管跺，因此我們?cè)谔卣飨蛄恐袨榈诙环峙湟弧?duì)于256位向量禾进，BRIEF然后在移動(dòng)到下一個(gè)關(guān)鍵點(diǎn)之前對(duì)相同的關(guān)鍵點(diǎn)重復(fù)此過(guò)程256次豁跑。然后將256像素亮度比較的結(jié)果放入該一個(gè)關(guān)鍵點(diǎn)的二進(jìn)制特征向量中。
ORB 比例和旋轉(zhuǎn)不變性
ORB使用FAST檢測(cè)圖像中的關(guān)鍵點(diǎn)泻云，它需要做幾個(gè)額外的步驟確保他可以檢測(cè)到物體艇拍，無(wú)論它們的大小或圖像中的位置。
- ORB算法從構(gòu)建圖像金字塔開(kāi)始
圖像金字塔是單個(gè)圖像的多尺度表示宠纯，其由一系列圖像組成卸夕，所有圖像都是不同分辨率的原始圖像的版本。
金字塔中的每個(gè)級(jí)別都包含上一級(jí)圖像的下采樣版本婆瓜。下采樣意味著圖像分辨率已降低快集。
在該示例中贡羔，圖像被下采樣兩倍，因此最初為四乘四平方區(qū)域的部分現(xiàn)在是兩乘二的方形區(qū)域个初。
圖像的下采樣版本包含像素減少乖寒，尺寸縮小了兩倍。在這里院溺，我們看到一個(gè)五層圖像金字塔的例子宵统。在每個(gè)級(jí)別，圖像被下采樣兩倍覆获，到第四級(jí)马澈，我們有一個(gè)圖像十六分之一原始圖像的分辨率。
一旦ORB創(chuàng)建了圖像金字塔弄息，它使用快速算法在每個(gè)級(jí)別快速定位不同大小的圖像中的關(guān)鍵點(diǎn)痊班。由于金字塔的每個(gè)級(jí)別由原始圖像的較小版本組成，因此原始圖像中的任何對(duì)象也將在金字塔的每個(gè)級(jí)別上減小尺寸摹量。因此涤伐，通過(guò)在每個(gè)級(jí)別定位關(guān)鍵點(diǎn)，ORB有效地定位不同比例的對(duì)象的關(guān)鍵點(diǎn)缨称。 以這種方式凝果，ORB部分尺度不變的。這非常重要睦尽，因?yàn)閷?duì)象不太可能在每個(gè)圖像中以完全相同的尺寸出現(xiàn)器净。
所以，ORB具有與該圖像金字塔的每個(gè)級(jí)別相關(guān)聯(lián)的關(guān)鍵點(diǎn)当凡。 -
在金字塔的所有層面的關(guān)鍵點(diǎn)都被找到后ORB現(xiàn)在為每個(gè)關(guān)鍵點(diǎn)指定方向山害。像左或右面對(duì)，取決于強(qiáng)度水平如何圍繞該關(guān)鍵點(diǎn)變化沿量。ORB將首先選擇金字塔零級(jí)的圖像浪慌。對(duì)于此圖像，它現(xiàn)在將計(jì)算其關(guān)鍵點(diǎn)的方向朴则。首先計(jì)算以關(guān)鍵點(diǎn)為中心的盒子內(nèi)的強(qiáng)度質(zhì)心权纤，強(qiáng)度質(zhì)心可以被認(rèn)為是給定貼片中的平均像素強(qiáng)度的位置。一旦計(jì)算出強(qiáng)度質(zhì)心乌妒，就可以通過(guò)將關(guān)鍵點(diǎn)的矢量繪制到強(qiáng)度質(zhì)心來(lái)獲得關(guān)鍵點(diǎn)的方向汹想，如此處所示。
該特定關(guān)鍵點(diǎn)的方向向下并朝向左側(cè)芥被，因?yàn)閳D像的該區(qū)域中的亮度在該方向上增加欧宜。一旦為金字塔零級(jí)圖像中的每個(gè)關(guān)鍵點(diǎn)指定了方向，ORB現(xiàn)在對(duì)所有其他金字塔等級(jí)的圖像重復(fù)相同的處理拴魄。值得注意的是，在圖像金字塔的每個(gè)級(jí)別上，補(bǔ)丁大小的大小都沒(méi)有減小匹中。因此夏漱，在金字塔的每個(gè)級(jí)別由相同補(bǔ)丁覆蓋的圖像區(qū)域?qū)⒏蟆?/p>
在此圖像中，圓圈表示每個(gè)關(guān)鍵點(diǎn)的大小顶捷。在金字塔的較高級(jí)別中找到具有較大尺寸的關(guān)鍵點(diǎn)挂绰。
- 找到并指定關(guān)鍵點(diǎn)的方向后，ORB現(xiàn)在使用修改后的brief來(lái)創(chuàng)建特征向量服赎，稱為rBRIEF葵蒂。無(wú)論對(duì)象的方向如何，都可以為關(guān)鍵點(diǎn)創(chuàng)建相同的矢量重虑。這使ORB算法旋轉(zhuǎn)不變践付，這意味著它可以檢測(cè)到以任何角度旋轉(zhuǎn)的圖像中的相同關(guān)鍵點(diǎn)。RBRIEF以與Brief相同的方式開(kāi)始缺厉，通過(guò)在給定關(guān)鍵點(diǎn)周圍的定義補(bǔ)丁內(nèi)選擇256個(gè)隨機(jī)像素對(duì)來(lái)構(gòu)造256位向量永高。然后它按照關(guān)鍵點(diǎn)的方向角，旋轉(zhuǎn)這些隨機(jī)像素對(duì)提针，以便將隨機(jī)點(diǎn)與關(guān)鍵點(diǎn)的方向?qū)R命爬。最后，rBRIEF比較隨機(jī)像素對(duì)的亮度辐脖，并相應(yīng)地分配1和0來(lái)創(chuàng)建相應(yīng)的特征向量饲宛。圖像中找到的所有關(guān)鍵點(diǎn)的所有特征向量的集合，被稱為ORB描述符
首先嗜价，我們將讀入圖像落萎，然后構(gòu)建并顯示圖像金字塔的幾個(gè)層
import numpy as np
import matplotlib.pyplot as plt
import cv2
%matplotlib inline
# Read in the image
image = cv2.imread('images/rainbow_flag.jpg')
# Change color to RGB (from BGR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
plt.imshow(image)
level_1 = cv2.pyrDown(image)
level_2 = cv2.pyrDown(level_1)
level_3 = cv2.pyrDown(level_2)
# Display the images
f, (ax1,ax2,ax3,ax4) = plt.subplots(1, 4, figsize=(20,10))
ax1.set_title('original')
ax1.imshow(image)
ax2.imshow(level_1)
ax2.set_xlim([0, image.shape[1]])
ax2.set_ylim([image.shape[0], 0])
ax3.imshow(level_2)
ax3.set_xlim([0, image.shape[1]])
ax3.set_ylim([image.shape[0], 0])
ax4.imshow(level_3)
ax4.set_xlim([0, image.shape[1]])
ax4.set_ylim([image.shape[0], 0])
視頻中ORB
ORB常用的一種， 是跟蹤和識(shí)別實(shí)時(shí)視頻流中的對(duì)象炭剪。 然后练链，對(duì)于傳入視頻流中的每個(gè)幀，我們計(jì)算ORB描述符奴拦，并使用匹配函數(shù)進(jìn)行比較媒鼓。 如果我們發(fā)現(xiàn)匹配函數(shù)返回超過(guò)某個(gè)匹配閾值的匹配數(shù)，我們就可以斷定對(duì)象在幀中错妖。網(wǎng)格閾值是一個(gè)可以設(shè)置的自由參數(shù)绿鸣。例如，如果特定對(duì)象的ORB描述符具有100個(gè)關(guān)鍵點(diǎn)暂氯，則可以將閾值設(shè)置為該特定對(duì)象的關(guān)鍵點(diǎn)數(shù)量的35％潮模，50％或90％。如果將閾值設(shè)置為35%痴施，則意味著描述該對(duì)象的100個(gè)關(guān)鍵點(diǎn)中至少有35個(gè)關(guān)鍵點(diǎn)必須匹配擎厢，以便說(shuō)明該對(duì)象在幀中究流。所有這些步驟都可以在近乎實(shí)時(shí)的情況下完成，因?yàn)镺RBS二進(jìn)制描述符动遭， 計(jì)算和比較非撤姨剑快。 當(dāng)您想要檢測(cè)具有許多不受圖像背景影響的一致特征的對(duì)象時(shí)厘惦，ORB算法工作得最好偷仿。例如，ORB可以很好地用于面部檢測(cè)宵蕉，因?yàn)槟槻坑泻芏嗵卣髟途玻热缪劬妥旖?無(wú)論人身在何處，這種情況都不會(huì)改變羡玛。
Oriented FAST and Rotated BRIEF (ORB)
- Loading Images and Importing Resources
cv2.imread（）函數(shù)將圖像加載為BGR别智，我們將圖像轉(zhuǎn)換為RGB，將BGR圖像轉(zhuǎn)換為灰度進(jìn)行分析缝左。
%matplotlib inline
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [20,10]
# Load the training image
image = cv2.imread('./images/face.jpeg')
# Convert the training image to RGB
training_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Convert the training image to gray Scale
training_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# Display the images
plt.subplot(121)
plt.title('Original Training Image')
plt.imshow(training_image)
plt.subplot(122)
plt.title('Gray Scale Training Image')
plt.imshow(training_gray, cmap = 'gray')
plt.show()
- 定位關(guān)鍵點(diǎn)
ORB算法的第一步是定位訓(xùn)練圖像中的所有關(guān)鍵點(diǎn)亿遂。 找到關(guān)鍵點(diǎn)后，ORB會(huì)創(chuàng)建相應(yīng)的二進(jìn)制特征向量渺杉，并在ORB描述符中將它們組合在一起蛇数。使用OpenCV的ORB類來(lái)定位關(guān)鍵點(diǎn)并創(chuàng)建相應(yīng)的ORB描述符。 使用ORB_create（）函數(shù)設(shè)置ORB算法的參數(shù)是越。 ORB_create（）函數(shù)的參數(shù)及其默認(rèn)值如下：
cv2.ORB_create(nfeatures = 500, scaleFactor = 1.2, nlevels = 8, edgeThreshold = 31, firstLevel = 0, WTA_K = 2, scoreType = HARRIS_SCORE, patchSize = 31, fastThreshold = 20)
參數(shù)：
-
nfeatures - int
確定要定位的最大特征數(shù)（關(guān)鍵點(diǎn)）耳舅。 -
scaleFactor - float
金字塔抽取率必須大于1. ORB使用圖像金字塔來(lái)查找要素，因此必須提供金字塔中每個(gè)圖層與金字塔所具有的級(jí)別數(shù)之間的比例因子倚评。scaleFactor = 2表示經(jīng)典金字塔浦徊，其中每個(gè)下一級(jí)的像素比前一級(jí)少4倍。大比例因子將減少發(fā)現(xiàn)特征數(shù)量天梧。 -
nlevels - int
金字塔等級(jí)的數(shù)量盔性，最小級(jí)別的線性大小等于input_image_linear_size / pow（scaleFactor，nlevels）呢岗。 -
edgeThreshold - - int
未檢測(cè)到要素的邊框大小冕香。 由于關(guān)鍵點(diǎn)具有特定的像素大小，因此必須從搜索中排除圖像的邊緣后豫。 邊緣閾值的大小應(yīng)等于或大于patchSize參數(shù)悉尾。 -
firstLevel - int
此參數(shù)允許您確定應(yīng)將哪個(gè)級(jí)別視為金字塔中的第一級(jí)別。 它在當(dāng)前實(shí)現(xiàn)中應(yīng)為0挫酿。 通常构眯，具有統(tǒng)一標(biāo)度的金字塔等級(jí)被認(rèn)為是第一級(jí)。 -
WTA_K - int
用于生成定向的BRIEF描述符的每個(gè)元素的隨機(jī)像素的數(shù)量早龟。 可能的值為2,3和4惫霸，其中2為默認(rèn)值猫缭。 例如，值3意味著一次選擇三個(gè)隨機(jī)像素來(lái)比較它們的亮度它褪。 返回最亮像素的索引饵骨。 由于有3個(gè)像素翘悉，因此返回的索引將為0,1或2茫打。 -
scoreType - int
此參數(shù)可以設(shè)置為HARRIS_SCORE或FAST_SCORE。 默認(rèn)的HARRIS_SCORE表示Harris角算法用于對(duì)要素進(jìn)行排名妖混。 該分?jǐn)?shù)僅用于保留最佳功能老赤。 FAST_SCORE生成的關(guān)鍵點(diǎn)稍差，但計(jì)算起來(lái)要快一些制市。 -
patchSize - int
面向BRIEF描述符使用的補(bǔ)丁的大小抬旺。 當(dāng)然，在較小的金字塔層上祥楣，由特征覆蓋的感知圖像區(qū)域?qū)⒏蟆?br> 我們可以看到开财，cv2。 ORB_create（）函數(shù)支持各種參數(shù)误褪。 前兩個(gè)參數(shù)（nfeatures和scaleFactor）可能是您最有可能改變的參數(shù)责鳍。 其他參數(shù)可以安全地保留其默認(rèn)值，您將獲得良好的結(jié)果兽间。
在下面的代碼中历葛，我們將使用ORB_create（）函數(shù)將我們想要檢測(cè)的關(guān)鍵點(diǎn)的最大數(shù)量設(shè)置為200，并將金字塔抽取率設(shè)置為2.1嘀略。 然后恤溶，我們將使用.detectAndCompute（image）方法定位給定訓(xùn)練圖像中的關(guān)鍵點(diǎn)并計(jì)算其對(duì)應(yīng)的ORB描述符。 最后帜羊，我們將使用cv2.drawKeypoints（）函數(shù)來(lái)可視化ORB算法找到的關(guān)鍵點(diǎn)咒程。
# Import copy to make copies of the training image
import copy
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(200, 2.0)
# Find the keypoints in the gray scale training image and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask.
keypoints, descriptor = orb.detectAndCompute(training_gray, None)
# Create copies of the training image to draw our keypoints on
keyp_without_size = copy.copy(training_image)
keyp_with_size = copy.copy(training_image)
# Draw the keypoints without size or orientation on one copy of the training image
cv2.drawKeypoints(training_image, keypoints, keyp_without_size, color = (0, 255, 0))
# Draw the keypoints with size and orientation on the other copy of the training image
cv2.drawKeypoints(training_image, keypoints, keyp_with_size, flags = cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# Display the image with the keypoints without size or orientation
plt.subplot(121)
plt.title('Keypoints Without Size or Orientation')
plt.imshow(keyp_without_size)
# Display the image with the keypoints with size and orientation
plt.subplot(122)
plt.title('Keypoints With Size and Orientation')
plt.imshow(keyp_with_size)
plt.show()
# Print the number of keypoints detected
print("\nNumber of keypoints Detected: ", len(keypoints))
正如我們?cè)谟覉D所示，每個(gè)關(guān)鍵點(diǎn)都有一個(gè)中心讼育，一個(gè)大小和一個(gè)角度帐姻。 中心確定圖像中每個(gè)關(guān)鍵點(diǎn)的位置; 每個(gè)關(guān)鍵點(diǎn)的大小由BRIEF用來(lái)創(chuàng)建其特征向量的補(bǔ)丁大小決定; 角度告訴我們由rBRIEF確定的關(guān)鍵點(diǎn)的方向。
特征匹配
一旦我們獲得了訓(xùn)練和查詢圖像的ORB描述符窥淆，最后一步就是使用相應(yīng)的ORB描述符在兩個(gè)圖像之間進(jìn)行關(guān)鍵點(diǎn)匹配卖宠。 這種匹配通常由匹配函數(shù)執(zhí)行。 最常用的匹配函數(shù)之一稱為Brute-Force忧饭。
在下面的代碼中扛伍，我們將使用OpenCV的BFMatcher類來(lái)比較訓(xùn)練和查詢圖像中的關(guān)鍵點(diǎn)。使用cv2.BFMatcher（）函數(shù)設(shè)置Brute-Force匹配器的參數(shù)词裤。 cv2.BFMatcher（）函數(shù)的參數(shù)及其默認(rèn)值如下：
-
normType
指定用于確定匹配質(zhì)量的度量刺洒。默認(rèn)情況下鳖宾，NoMyType＝CV2..NoqILL2，它測(cè)量?jī)蓚€(gè)描述符之間的距離逆航。然而鼎文，對(duì)于二進(jìn)制描述符，如由ORB創(chuàng)建的因俐，漢明度量（Hamming metric）更適合拇惋。漢明度量通過(guò)計(jì)算二進(jìn)制描述符之間的不同位的數(shù)目來(lái)確定距離。當(dāng)使用WTAYK＝2創(chuàng)建ORB描述符時(shí)抹剩，選擇兩個(gè)隨機(jī)像素并在亮度上進(jìn)行比較撑帖。最亮像素的索引以0或1返回。這樣的輸出只占用1位澳眷，因此應(yīng)該使用CV2..NojyHaMin度量胡嘿。另一方面，如果使用WTA_K=3創(chuàng)建ORB描述符钳踊，則選擇三個(gè)隨機(jī)像素并在亮度方面進(jìn)行比較衷敌。最亮像素的索引以0, 1或2的形式返回。這種輸出將占用2比特拓瞪，因此應(yīng)該使用漢明距離的特殊變體缴罗，稱為cv2.NORM_HAMMING2（2代表2比特）。然后吴藻，對(duì)于所選擇的任何度量瞒爬，當(dāng)比較訓(xùn)練和查詢圖像中的關(guān)鍵點(diǎn)時(shí)，認(rèn)為度量較泄当ぁ（它們之間的距離）的對(duì)是最佳匹配侧但。 -
crossCheck - bool
一個(gè)布爾變量，可以設(shè)置為“True”或“False”航罗。 交叉檢查對(duì)于消除錯(cuò)誤匹配非常有用禀横。 交叉檢查通過(guò)執(zhí)行兩次匹配過(guò)程來(lái)工作。 第一次將訓(xùn)練圖像中的關(guān)鍵點(diǎn)與查詢圖像中的關(guān)鍵點(diǎn)進(jìn)行比較; 然而粥血，第二次柏锄，將查詢圖像中的關(guān)鍵點(diǎn)與訓(xùn)練圖像中的關(guān)鍵點(diǎn)進(jìn)行比較（即比較向后進(jìn)行）。 啟用交叉檢查時(shí)复亏，只有當(dāng)訓(xùn)練圖像中的關(guān)鍵點(diǎn)A與查詢圖像中的關(guān)鍵點(diǎn)B最佳匹配時(shí)趾娃，匹配才被視為有效，反之亦然（即缔御，如果查詢中的關(guān)鍵點(diǎn)B 圖像是訓(xùn)練圖像中點(diǎn)A的最佳匹配抬闷。
一旦設(shè)置了* BFMatcher *的參數(shù)，我們就可以使用.match（descriptors_train，descriptors_query）方法笤成，使用它們的ORB描述符找到訓(xùn)練圖像和查詢圖像之間的匹配關(guān)鍵點(diǎn)评架。 最后，我們將使用cv2.drawMatches（）函數(shù)來(lái)可視化Brute-Force匹配器找到的匹配關(guān)鍵點(diǎn)炕泳。 此函數(shù)水平堆疊訓(xùn)練和查詢圖像纵诞，并將訓(xùn)練圖像中關(guān)鍵點(diǎn)的線條繪制到查詢圖像中對(duì)應(yīng)的最佳匹配關(guān)鍵點(diǎn)。 請(qǐng)記住培遵，為了更清楚地查看ORB算法的屬性浙芙，在以下示例中，我們將使用與我們的訓(xùn)練和查詢圖像相同的圖像荤懂。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the query image
image2 = cv2.imread('./images/face.jpeg')
# Convert the training image to RGB
training_image = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
# Convert the query image to RGB
query_image = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
# Display the training and query images
plt.subplot(121)
plt.title('Training Image')
plt.imshow(training_image)
plt.subplot(122)
plt.title('Query Image')
plt.imshow(query_image)
plt.show()
# Convert the training image to gray scale
training_gray = cv2.cvtColor(training_image, cv2.COLOR_BGR2GRAY)
# Convert the query image to gray scale
query_gray = cv2.cvtColor(query_image, cv2.COLOR_BGR2GRAY)
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(1000, 2.0)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create a Brute Force Matcher object. Set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 300 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:300], query_gray, flags = 2)
# Display the best matching points
plt.title('Best Matching Points')
plt.imshow(result)
plt.show()
# Print the number of keypoints detected in the training image
print("Number of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching points between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
ORB 主要屬性
- 比例不變性
- 旋轉(zhuǎn)不變性
- 照明不變性
- 噪音不變性
Scale Invariance
ORB算法是尺度不變的茁裙。 這意味著它能夠檢測(cè)圖像中的對(duì)象塘砸，無(wú)論其大小如何节仿。 為了看到這一點(diǎn)，我們現(xiàn)在將使用我們的蠻力匹配器來(lái)匹配訓(xùn)練圖像和原始訓(xùn)練圖像大小的1/4的查詢圖像之間的點(diǎn)掉蔬。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the query image
image2 = cv2.imread('./images/faceQS.png')
# Convert the training image to RGB
training_image = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
# Convert the query image to RGB
query_image = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
# Display the images
plt.subplot(121)
plt.title('Training Image')
plt.imshow(training_image)
plt.subplot(122)
plt.title('Query Image')
plt.imshow(query_image)
plt.show()
# Convert the training image to gray scale
training_gray = cv2.cvtColor(training_image, cv2.COLOR_BGR2GRAY)
# Convert the query image to gray scale
query_gray = cv2.cvtColor(query_image, cv2.COLOR_BGR2GRAY)
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(1000, 2.0)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create a Brute Force Matcher object. Set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 30 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:30], query_gray, flags = 2)
# Display the best matching points
plt.title('Best Matching Points')
plt.imshow(result)
plt.show()
# Print the shape of the training image
print('\nThe Training Image has shape:', training_gray.shape)
#Print the shape of the query image
print('The Query Image has shape:', query_gray.shape)
# Print the number of keypoints detected in the training image
print("\nNumber of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching points between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
在上面的示例中廊宪，請(qǐng)注意訓(xùn)練圖像是553 x 471像素，而查詢圖像是138 x 117像素女轿，是原始訓(xùn)練圖像大小的1/4箭启。 另請(qǐng)注意，查詢圖像中檢測(cè)到的關(guān)鍵點(diǎn)數(shù)量?jī)H為65蛉迹，遠(yuǎn)小于訓(xùn)練圖像中的831個(gè)關(guān)鍵點(diǎn)傅寡。 然而，我們可以看到我們的蠻力匹配器可以將查詢圖像中的大多數(shù)關(guān)鍵點(diǎn)與訓(xùn)練圖像中的相應(yīng)關(guān)鍵點(diǎn)進(jìn)行匹配北救。
Rotational Invariance
ORB算法也是旋轉(zhuǎn)不變的荐操。 這意味著無(wú)論方向如何，它都能夠檢測(cè)圖像中的對(duì)象珍策。 為了看到這一點(diǎn)托启，我們現(xiàn)在將使用我們的Brute-Force匹配器來(lái)匹配訓(xùn)練圖像和旋轉(zhuǎn)了90度的查詢圖像之間的點(diǎn)。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the query image
image2 = cv2.imread('./images/faceR.jpeg')
# Convert the training image to RGB
training_image = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
# Convert the query image to RGB
query_image = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
# Display the images
plt.subplot(121)
plt.title('Training Image')
plt.imshow(training_image)
plt.subplot(122)
plt.title('Query Image')
plt.imshow(query_image)
plt.show()
# Convert the training image to gray scale
training_gray = cv2.cvtColor(training_image, cv2.COLOR_BGR2GRAY)
# Convert the query image to gray scale
query_gray = cv2.cvtColor(query_image, cv2.COLOR_BGR2GRAY)
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(1000, 2.0)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create a Brute Force Matcher object. Set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 100 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:100], query_gray, flags = 2)
# Display the best matching points
plt.title('Best Matching Points')
plt.imshow(result)
plt.show()
# Print the number of keypoints detected in the training image
print("\nNumber of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching points between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
在上面的例子中攘宙，我們看到在兩個(gè)圖像中檢測(cè)到的關(guān)鍵點(diǎn)的數(shù)量非常相似屯耸，即使查詢圖像被旋轉(zhuǎn)，我們的Brute-Force匹配器仍然可以匹配找到的關(guān)鍵點(diǎn)的大約78％蹭劈。 此外疗绣，請(qǐng)注意大多數(shù)匹配關(guān)鍵點(diǎn)都接近特定的面部特征，例如眼睛铺韧，鼻子和嘴巴多矮。
Illumination Invariance
ORB算法也是照明不變的。 這意味著無(wú)論照明如何祟蚀，它都能夠檢測(cè)圖像中的物體工窍。 為了看到這一點(diǎn)割卖，我們現(xiàn)在將使用我們的Brute-Force匹配器來(lái)匹配訓(xùn)練圖像和更亮的查詢圖像之間的點(diǎn)。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the query image
image2 = cv2.imread('./images/faceRI.png')
# Convert the training image to RGB
training_image = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
# Convert the query image to RGB
query_image = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
# Display the images
plt.subplot(121)
plt.title('Training Image')
plt.imshow(training_image)
plt.subplot(122)
plt.title('Query Image')
plt.imshow(query_image)
plt.show()
# Convert the training image to gray scale
training_gray = cv2.cvtColor(training_image, cv2.COLOR_BGR2GRAY)
# Convert the query image to gray scale
query_gray = cv2.cvtColor(query_image, cv2.COLOR_BGR2GRAY)
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(1000, 2.0)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create a Brute Force Matcher object. Set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 100 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:100], query_gray, flags = 2)
# Display the best matching points
plt.title('Best Matching Points')
plt.imshow(result)
plt.show()
# Print the number of keypoints detected in the training image
print("\nNumber of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching points between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
在上面的例子中患雏，我們看到在兩個(gè)圖像中檢測(cè)到的關(guān)鍵點(diǎn)的數(shù)量再次非常相似鹏溯，即使查詢圖像更亮，我們的Brute-Force匹配器仍然可以匹配找到的關(guān)鍵點(diǎn)的大約63％淹仑。
Noise Invariance
ORB算法也是噪聲不變的丙挽。 這意味著它能夠檢測(cè)圖像中的對(duì)象，即使圖像具有一定程度的噪聲匀借。 為了看到這一點(diǎn)颜阐，我們現(xiàn)在將使用我們的Brute-Force匹配器來(lái)匹配訓(xùn)練圖像和具有大量噪聲的查詢圖像之間的點(diǎn)。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the noisy, gray scale query image.
image2 = cv2.imread('./images/faceRN5.png')
# Convert the query image to gray scale
query_gray = cv2.cvtColor(image2, cv2.COLOR_BGR2GRAY)
# Convert the training image to gray scale
training_gray = cv2.cvtColor(image1, cv2.COLOR_BGR2GRAY)
# Display the images
plt.subplot(121)
plt.imshow(training_gray, cmap = 'gray')
plt.title('Gray Scale Training Image')
plt.subplot(122)
plt.imshow(query_gray, cmap = 'gray')
plt.title('Query Image')
plt.show()
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(1000, 1.3)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create a Brute Force Matcher object. We set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 100 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:100], query_gray, flags = 2)
# we display the image
plt.title('Best Matching Points')
plt.imshow(result)
plt.show()
# Print the number of keypoints detected in the training image
print("Number of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching points between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
在上面的例子中吓肋，我們?cè)俅慰吹絻蓚€(gè)圖像中檢測(cè)到的關(guān)鍵點(diǎn)的數(shù)量非常相似凳怨，即使查詢圖像有很多噪聲，我們的Brute-Force匹配器仍然可以找到大約63％的關(guān)鍵點(diǎn)是鬼。 此外肤舞，請(qǐng)注意大多數(shù)匹配關(guān)鍵點(diǎn)都接近特定的面部特征，例如眼睛均蜜，鼻子和嘴巴李剖。 此外，我們可以看到有一些功能不太匹配囤耳，但可能因?yàn)閳D像區(qū)域中的強(qiáng)度模式相似而被選中篙顺。 我們還要指出，在這種情況下充择，我們使用的金字塔抽取率為1.3德玫，而不是前面示例中使用的值2.0，因?yàn)樵谶@種特殊情況下聪铺，它會(huì)產(chǎn)生更好的結(jié)果化焕。
目標(biāo)檢測(cè)
實(shí)現(xiàn)ORB算法以在另一圖像中檢測(cè)訓(xùn)練圖像中的面部拇涤。 像往常一樣矢门，我們將首先加載我們的培訓(xùn)和查詢圖像。
import cv2
import matplotlib.pyplot as plt
# Set the default figure size
plt.rcParams['figure.figsize'] = [14.0, 7.0]
# Load the training image
image1 = cv2.imread('./images/face.jpeg')
# Load the query image
image2 = cv2.imread('./images/Team.jpeg')
# Convert the training image to RGB
training_image = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
# Convert the query image to RGB
query_image = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
# Display the images
plt.subplot(121)
plt.imshow(training_image)
plt.title('Training Image')
plt.subplot(122)
plt.imshow(query_image)
plt.title('Query Image')
plt.show()
在這種特定情況下墨坚，訓(xùn)練圖像包含面部键兜，因此凤类，如我們所見(jiàn)，檢測(cè)到的大多數(shù)關(guān)鍵點(diǎn)接近面部特征普气，例如眼睛谜疤，鼻子和嘴巴。 另一方面，我們的查詢圖像是一組人的圖片夷磕，其中一個(gè)是我們想要檢測(cè)的女人履肃。 現(xiàn)在讓我們檢測(cè)查詢圖像的關(guān)鍵點(diǎn)。
# Set the default figure size
plt.rcParams['figure.figsize'] = [34.0, 34.0]
# Convert the training image to gray scale
training_gray = cv2.cvtColor(training_image, cv2.COLOR_BGR2GRAY)
# Convert the query image to gray scale
query_gray = cv2.cvtColor(query_image, cv2.COLOR_BGR2GRAY)
# Set the parameters of the ORB algorithm by specifying the maximum number of keypoints to locate and
# the pyramid decimation ratio
orb = cv2.ORB_create(5000, 2.0)
# Find the keypoints in the gray scale training and query images and compute their ORB descriptor.
# The None parameter is needed to indicate that we are not using a mask in either case.
keypoints_train, descriptors_train = orb.detectAndCompute(training_gray, None)
keypoints_query, descriptors_query = orb.detectAndCompute(query_gray, None)
# Create copies of the query images to draw our keypoints on
query_img_keyp = copy.copy(query_image)
# Draw the keypoints with size and orientation on the copy of the query image
cv2.drawKeypoints(query_image, keypoints_query, query_img_keyp, flags = cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# Display the query image with the keypoints with size and orientation
plt.title('Keypoints With Size and Orientation', fontsize = 30)
plt.imshow(query_img_keyp)
plt.show()
# Print the number of keypoints detected
print("\nNumber of keypoints Detected: ", len(keypoints_query))
我們可以看到查詢圖像在圖像的許多部分都有關(guān)鍵點(diǎn)坐桩。 現(xiàn)在我們已經(jīng)獲得了訓(xùn)練圖像和查詢圖像的關(guān)鍵點(diǎn)和ORB描述符尺棋，我們可以使用Brute-Force匹配器來(lái)嘗試在查詢圖像中找到女性的臉部。
# Set the default figure size
plt.rcParams['figure.figsize'] = [34.0, 34.0]
# Create a Brute Force Matcher object. We set crossCheck to True so that the BFMatcher will only return consistent
# pairs. Such technique usually produces best results with minimal number of outliers when there are enough matches.
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
# Perform the matching between the ORB descriptors of the training image and the query image
matches = bf.match(descriptors_train, descriptors_query)
# The matches with shorter distance are the ones we want. So, we sort the matches according to distance
matches = sorted(matches, key = lambda x : x.distance)
# Connect the keypoints in the training image with their best matching keypoints in the query image.
# The best matches correspond to the first elements in the sorted matches list, since they are the ones
# with the shorter distance. We draw the first 85 mathces and use flags = 2 to plot the matching keypoints
# without size or orientation.
result = cv2.drawMatches(training_gray, keypoints_train, query_gray, keypoints_query, matches[:85], query_gray, flags = 2)
# we display the image
plt.title('Best Matching Points', fontsize = 30)
plt.imshow(result)
plt.show()
# Print the number of keypoints detected in the training image
print("Number of Keypoints Detected In The Training Image: ", len(keypoints_train))
# Print the number of keypoints detected in the query image
print("Number of Keypoints Detected In The Query Image: ", len(keypoints_query))
# Print total number of matching Keypoints between the training and query images
print("\nNumber of Matching Keypoints Between The Training and Query Images: ", len(matches))
我們可以清楚地看到绵跷，即使查詢圖像中有許多面部和對(duì)象膘螟，我們的Brute-Force匹配器也能夠在查詢圖像中正確定位女性的面部。
Histogram of Oriented Gradients （HOG）方向梯度直方圖
直方圖(Histogram)
直方圖就是數(shù)據(jù)分布的一種圖形表現(xiàn)
以灰度圖像為例:
假設(shè)你想繪制出這張薄餅圖的強(qiáng)度數(shù)據(jù)直方圖,像素值的范圍在 0 到 255 之間,所以我們可以把這些值分成若干組,創(chuàng)建 32 個(gè)組 每組包含 8 個(gè)像素值 所以第一組范圍是 0 到 7碾局、然后 8 到 15 以此類推一直到 248 到 255
方向梯度
方向很簡(jiǎn)單 指的就是圖像梯度的方向或朝向荆残，用 Sobel 算子來(lái)計(jì)算梯度的幅值和方向
把三個(gè)術(shù)語(yǔ)結(jié)合在一起，HOG 就是指一張有關(guān)圖像梯度方向的直方圖
- 首先 HOG 會(huì)接受一張圖像净当，然后計(jì)算每個(gè)像素的梯度幅值和方向
- HOG 實(shí)際上會(huì)把這些像素分成若干個(gè)較大的正方形單元 單元大小通常是 8 乘 8 如果圖片小一些 單元也就小一些内斯，如果是 8 乘 8 的單元，那就有 64 個(gè)梯度值蚯瞧，HOG 會(huì)計(jì)算每個(gè)單元相同的梯度方向有多少嘿期，將這些梯度的幅值相加，得到梯度強(qiáng)度
-
接著 HOG 會(huì)把所有方向數(shù)據(jù)放到直方圖里埋合，這個(gè)直方圖有九組 也就是九個(gè)值范圍，不過(guò)你可以建立更多組來(lái)進(jìn)一步分類數(shù)據(jù)
