霍夫空間轉換與互補視覺資訊運作結構下之規律重複性服飾圖案偵測

Abstract

在日常生活中，人們的衣著服飾圖案形形色色，而其中重複性規律排列之圖案佔有相當的比率，例如:方格紋、千鳥格紋、平行條紋，大小圓花等等，均為極其普遍常見的重複性圖案。 一般而言，二維重複性規律排列圖案乃由一基本構形(圓，方格，長條或更複雜之紋案)延特定軌跡依一定間距重複繪製，再將此單一軌跡依特定規律展延為二維分布而形成。在此前提之下，規律排列之諸多基本構形之間，不難發現分屬相異基本構形邊緣之諸像素具有共線之特點。 基於前述觀察，本論文以霍夫轉換可偵測影像中之共線點之特性為基礎，嘗試找出影像中出現重複性規則服飾圖案的區域。在由測試影像取得四組特定角度之邊緣特徵後(水平與垂直、45與135度角)， 利用霍夫轉換及正交邊緣特徵分布上之互補關係，可大幅濾除與重複性規則服飾圖案無關的邊緣線段，從而使目標區域明確。 經實驗觀察，測試影像中重複性規律排列之圖案可由本論文實做之系統偵測出來，驗證出霍夫轉換可以應用在偵測重複性規律排列之圖案。

As can be seen in people’s daily wearing, repetitive fabric patterns are so common that simply cannot be shunted in one’s sights.

Considering the ubiquitous skewness of repetitive fabric patterns in the test images due to fabric wrinkles, in this thesis the term “repetitive fabric patterns” refers to, in the visual sense, the 2D distribution of a certain primitive pattern (which may be a dot/circle, a square, a strip, or a figure of more delicacy) along a trajectory at more or less regular distances, over a set of somewhat regularly-spaced trajectories. It should be noted that the distribution described here, and handled in the implementation to be detailed later, is by no means with mathematic strictness. In such a context, the co-linear phenomenon of pixels on boundary edges of different individual primitives is readily observed, and Hough transform is thus exploited for the detecting purpose. The detecting process starts with oriented edges extraction, by which 4 maps of edge feature are derived, respectively along vertical, horizontal, 45-degree, and135-degree diagonal directions. Hough transform is then applied, in cooperation with the use of complimentary information offered in the orthogonal sets of edge structures, to remove in a great extent the edges that are irrelevant to the target regions where repetitive patterns may potentially exist. A system is implemented and tested against a number of images of repetitive fabric patterns with variations in several aspects of imagery natures: 1. contrast of the primitive against the fabric, 2. scales of the primitive against primitive spacing, 3. types of primitives, by which the performance and limits of the system is therefore discerned.