《计算机专业英语》chapter11 Content-based Image Retrieval（CBIR)

Chapter 11 Content-based Image Retrieval（CBIR)Think About…What is cont
ent-based image retrieval? What does the term “content” mean?What
are low level image retrieval, region based image retrieval, and
semantic image retrieval respectively?From the historic overview
, how has CBIR evolved?What does it mean by multimedia informatio
n retrieval ? What Research Areas Are Involved In? Computer visi
on, pattern recognition, image processing, data mining, machine l
earning, human-computer interaction, artificial intelligenceAppli
cation: digital museum/libraries, safety of society, image/video
copy detection, GIS, medicine, education, entertainment, WWW, to
name just a fewDigital Image RetrievalSearch for digital images i
n large databasesFirst generation: laborious, subjectiveMetadata
(captions or keywords) →Image Second generation (content-based):
objectiveImage contents → Image Current way (semantic): subjectiv
e + objectiveImage contents + semantic feature → Image Our way: I
mage contents + semantic feature + keywords/captions → Image Sema
ntic gapWhat Is Content-based Image Retrieval“Content-based” mean
s that the search will analyze the actual contents of the image.
The term “content” in this context might refer colors, shapes, te
xtures, or any other information that can be derived from the ima
ge itselfLow Level Image Retrieval ColorExamining images based on
the colors they contain is one of the most widely used technique
s because it does not depend on image size or orientation. Color
searches will usually involve comparing color histograms, though
this is not the only technique in practiceColor SpaceRGBLightness
(亮度，即明暗）Hue（色调，即光的颜色）Saturation（饱和度，即颜色的深浅）Chrominance (色度）L
ow Level Image RetrievalShapeShape does not refer to the shape of
an image but to the shape of a particular region that is being s
ought out. Shapes will often be determined first applying segment
ation or edge detection to an image. In some cases accurate shape
detection will require human intervention because methods like s
egmentation are very difficult to completely automate.Edge Detect
ion Low Level Image RetrievalTextureTexture measures look for vis
ual patterns in images and how they are spatially defined. Textur
es are represented by texels which are then placed into a number
of sets, depending on how many textures are detected in the image
. These sets not only define the texture, but also where in the i
mage the texture is locatedTextureTextureCoarseness (粗糙度）Contrast
(对比度）Directionality(方向度）Linearity(线性度）Regularity(规整度）Roughness(粗
略度）Region Based Image Retrieval Semantic Image RetrievalHuman jud
gement of image similarity is subjective. Therefore latest resear
ch focus on deriving semantic features using machine learning tec
hniques to narrow down the semantic gapSemantic ModelLow level fe
atures (mixture of color, shape, texture, e.g. red circle)Objects
, e.g. a manSpatial relationship between the objects, e.g. a man
in front of the houseEnvironment, e.g. sandAction, e.g. runFeelin
g, e.g. happySemantic ModelLow level features: color, shape…Objec
ts: person, ball… Spatial relationship: person’s positions…Enviro
nment: sand, blue sky…Action: play volleyballFeeling: relax, happ
y…Semantic Feature RetrievalAutomatic image annotationBuild a sem
antic spaceContent-based Image RetrievalFrom the above historic o
verview, it can be seen how CBIR has evolved from low level image
retrieval to region based image retrieval and to semantic image
retrievalDatabasesStructure of CBIRDigital imagesFeature extracti
onUserQuery interfaceSearch engineImage databaseFeature databaseK
nowledge databaseImage Retrieval MethodsQuery by external pictori
al exampleBrand search, finger mark searchQuery by internal picto
rial exampleQuery by sketchKeywords/captionsCombination of the ab
ove methods Evaluation d bc
aPrecisionRecallPVREffectivenessEfficiencyFlexibilityA User Int
erface of Image Retrieval Future DirectionBeyond text, audio, images, and video, there has been significant recent interest in new media such as 3D modelsThe search for knowledge in all its forms is called Multimedia Information Retrieval (MIR)