\u8ad6\u6587\u767a\u8868\u4f1a\u958b\u50ac
\n\u65e5\u3000\u3000\u6642:\u4ee4\u548c2\u5e743\u670816\u65e5(\u6708)9:00\uff5e10:30
\n\u4f1a\u3000\u3000\u5834:\u77f3\u5ddd\u53f04\u53f7\u9928\u5730\u968e\u4f1a\u8b70\u5ba4(B04\/05\u53f7\u5ba4)
\n\u53f8\u4f1a\u6559\u54e1:\u5c71\u4e0b\u5e78\u5f66\u6559\u6388<\/p>\n
\u6c0f\u3000\u3000\u540d:\u30c1\u30e7\u30a6\u3000\u30ab\u30c4\u30b7<\/p>\n
\u8ad6\u6587\u984c\u76ee<\/p>\n
\u300cCriterion for image matching with global projection transformation
\ncorrelation and its acceleration algorithms\uff08\u5927\u5c40\u7684\u5c04\u5f71\u5909\u63db\u76f8\u95a2\u306b\u3088\u308b\u753b\u50cf\u30de\u30c3\u30c1
\n\u30f3\u30b0\u306e\u305f\u3081\u306e\u8a55\u4fa1\u57fa\u6e96\u304a\u3088\u3073\u305d\u306e\u9ad8\u901f\u8a08\u7b97\uff09\u300d<\/p>\n
\u6982\u8981<\/p>\n
Image matching is a strong ally in computer vision field, because it outputs
\nthe correspondence between the input images and templates. The most important
\nthing in image matching is the invariant matching result against adverse
\nconditions such as noise, color of background, geometrical transformations,
\netc. Furthermore, short calculation time is also highly expected for real-world
\napplications. This thesis proposes a theoretical criterion of GPT (Global
\nProjection Transformation) correlation matching which revises the deformations
\nbetween input images and templates by 2D projection transformation. In the
\nthesis, we also discuss the local features and acceleration algorithm to
\nreinforce the stability of the proposed criterion and shorten the calculation
\ntime.<\/p>\n
GPT correlation matching technique is originated by Wakahara and Yamashita to
\nrealize a projection-invariant image matching technique. However, theoretically,
\nthere is a serious defect in the conventional criterion. So firstly, we
\nintroduce the conventional GPT correlation matching model and point out a
\nproblem that the conventional approach does not guarantee the conservation of
\nthe L2 norm of the transformed images. The problem will lead mismatches during
\nthe optimization of the criterion. To solve this problem the theoretical
\ncriterion is proposed by introducing a norm normalization factor. Furthermore,
\nthe conventional GPT correlation matching realizes the 2D projection
\ntransformation by the alternative calculation of AT (Affine Transformation) and
\nPPT (Partial Projection Transformation), which causes the incompatibility during
\nthe matching process. This thesis also focuses on the norm normalization of the
\nsimultaneous calculation of AT and PPT. The mathematical derivation and
\nimplementation algorithm of the proposed criteria are described in detail and
\nsome demonstrations using toy image data are also illustrated. From the
\ndemonstrations, the proposed approach achieves not only higher correlations
\nbetween the input images and template, but also faster convergence speed.<\/p>\n
During the matching process, both the conventional approach and the proposed
\napproach use only the quantized gradient direction of the target pixel as a
\nkind of space filter to mitigate mismatches. However, in the field of natural
\nimage matching, the quantized gradient direction is greatly affected by
\ndeformations or distortion of the image. To achieve a more robust and stable
\nmatching against bad conditions of the images, this thesis also proposes some
\nkinds of local features of the histogram of oriented quantized gradient
\ndirection (HOG) information of a 5-by-5 block around each target pixel. For
\nexample, we simplify the HOG features by eliminating the non-significant
\ndirections. Or, we also perform k-means clustering to simplify the HOT features.
\nThe comparison among matching abilities using these HOG patterns is shown by
\ndemonstrations. From the demonstrations, in case the test images suffer from
\nlarge deformation, the matching results are stable with the assist of the local
\nfeatures, while the conventional space filter does not lead to the converge
\npoint any longer.<\/p>\n
Furthermore, the proposed criterion is expected to be utilized into real-world
\napplications, so short calculation time is highly required. This thesis
\nintroduces the acceleration algorithm of the proposed algorithm. The key idea
\nof the acceleration algorithm is to create two kinds of reference tables. These
\nreference tables can avoid the na\u00efve search of the image domain, which
\ndramatically reduce the calculation time to about one hundredth. Since the
\ncalculation time for creating the reference tables is still the fourth-order of
\nthe image size, an acceleration algorithm for creating the reference tables
\nusing integral images is also proposed. The same demonstrations as above-
\nmentioned section are conducted firstly to show the agreement of the matching
\nresults no matter the acceleration algorithm is used or not, and as well as the
\nsignificant reduction of the calculation time than the standard algorithm about
\none thousandth.<\/p>\n
Several experiments are performed to show the advantages of the proposed
\napproach. The experiments can be divided into two sections, one is the image
\nmatching part and the other is the image discrimination part. In image matching
\npart we perform “whole-to-whole” and “whole-to-part” template matchings using
\nnatural images from some famous datasets vis the conventional approaches, the
\nproposed approach as well as the acceleration algorithm, and the state-of-the-
\nart methods. In image discrimination part, the image discrimination of
\nhandwritten digit from MNIST database is performed vis the conventional
\napproaches and the accelerated proposed approach. The comparison of the
\ncorrelation values and the error rates output by these approaches are listed up.<\/p>\n
From the demonstrations and experiments, we can distinguish the outstanding
\nperformance of the proposed approach not only in image matching field but also
\nin image recognition field. It also has high potential to be applied to some
\nother real-world computer vision tasks.<\/p>\n","protected":false},"excerpt":{"rendered":"
\u8ad6\u6587\u767a\u8868\u4f1a\u958b\u50ac \u65e5\u3000\u3000\u6642:\u4ee4\u548c2\u5e743\u670816\u65e5(\u6708)9:00\uff5e10:30 \u4f1a\u3000\u3000\u5834:\u77f3\u5ddd\u53f04\u53f7\u9928\u5730\u968e\u4f1a\u8b70\u5ba4(B04\/05\u53f7\u5ba4) \u53f8\u4f1a\u6559\u54e1:\u5c71\u4e0b\u5e78\u5f66\u6559\u6388 \u6c0f\u3000\u3000\u540d:\u30c1\u30e7\u30a6\u3000\u30ab\u30c4\u30b7 \u8ad6\u6587\u984c\u76ee \u300cCriterion for im […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[],"class_list":["post-4524","post","type-post","status-publish","format-standard","hentry","category-presentations"],"_links":{"self":[{"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/posts\/4524","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/comments?post=4524"}],"version-history":[{"count":4,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/posts\/4524\/revisions"}],"predecessor-version":[{"id":4529,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/posts\/4524\/revisions\/4529"}],"wp:attachment":[{"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/media?parent=4524"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/categories?post=4524"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.ide.titech.ac.jp\/ja\/wp-json\/wp\/v2\/tags?post=4524"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}