# US 2017/0255839 A1 · Apparatus and method for extracting salient line for information sign

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2017/0255839 A1
- Document ID: US20170255839A1
- Status: Published application
- Kind: A1
- Application number: 15/365,630
- Filed: 2016-11-30
- Published: 2017-09-07
- Earliest claimed: 2016-03-03
- Assignees: Electronics and Telecommunications Research Institute
- Applicants: Electronics and Telecommunications Research Institute
- Inventors: Seung Jun HAN; Jeong Dan CHOI; Sang Heon PARK
- Web page: https://patentgrep.com/patent/US20170255839A1

## CPC

- G06V 20/582: {of traffic signs}
- G06T 5/80: Geometric correction
- G06T 7/68: of symmetry
- G06V 10/443: {by matching or filtering}
- G06V 10/462: {Salient features, e.g. scale invariant feature transforms [SIFT]}
- G06V 10/50: by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V 20/63: {Scene text, e.g. street names}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

Provided are an apparatus and method for extracting a salient line for an information sign. The apparatus includes an image input portion configured to receive an image including information indication lines to be detected, a signal calculator configured to generate a symmetric function signal consistent with a thickness and a direction of an information indication line having a preset thickness in the received image and perform a convolution of the information indication line and the generated symmetric function signal, a highest value enhancement portion configured to enhance highest values in results of the convolution using a signal quality improvement filter, and a local highest value extractor configured to find local highest values among the enhanced highest values and extract a line connecting the local highest values as a salient line connecting center points in the information indication line.

## Claims

### Claim 1

An apparatus for extracting a salient line for an information sign, the apparatus comprising:
an image input portion configured to receive an image including information indication lines to be detected;
a signal calculator configured to generate a symmetric function signal consistent with a thickness and a direction of an information indication line having a preset thickness in the received image and perform a convolution of the information indication line and the generated symmetric function signal;
a highest value enhancement portion configured to enhance highest values in results of the convolution using a signal quality improvement filter; and
a local highest value extractor configured to find local highest values among the enhanced highest values and extract a line connecting the local highest values as a salient line connecting center points in the information indication line.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the signal calculator generates a disk-shaped symmetric function signal consistent with an arbitrary information indication line having the preset thickness in all directions, and performs a convolution of the information indication line and the generated disk-shaped symmetric function signal.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the signal calculator generates a pillbox kernel which is a disk-shaped symmetric function defined as a function of image coordinates based on a center of the kernel and a radius of the kernel, and performs a convolution of the information indication line and the generated pillbox kernel.

### Claim 4 (depends on 3)

The apparatus of claim 3, wherein the signal calculator approximates the generated pillbox kernel to a value obtained by dividing a sum of image data corresponding to a circular region of the pillbox kernel by an area of the circular region, and performs a convolution of the information indication line and the approximated pillbox kernel.

### Claim 5 (depends on 3)

The apparatus of claim 3, wherein the signal calculator approximates a circular region of the generated pillbox kernel to a sum of a plurality of rectangular regions, and calculates the sum of the plurality of rectangular regions from sums of an integral image.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the signal calculator calculates the sum of the plurality of rectangular regions from the sums of the integral image using values at vertices in the integral image.

### Claim 7 (depends on 5)

The apparatus of claim 5, wherein the signal calculator performs a convolution of the information indication line and the approximated pillbox kernel, and calculates a parameter minimizing an error function of areas of divided arcs and areas of divided rectangular regions using least-squares optimization.

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein the highest value enhancement portion enhances the highest values of the convolution results using a morphological extrema filter (MEF) which is a signal quality improvement filter, and
the MEF is a sparse kernel having a plurality of pairs and is defined as a function of a filter gain and a preset threshold value.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein the local highest value extractor finds the local highest values in the enhanced highest values using non-maximum suppression (NMS) with respect to cross sections of the information indication line taken along lines perpendicular to the information indication line and extracts a line connecting the local highest values as the salient line connecting the center points in the information indication line.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the local highest value extractor finds the local highest values in the enhanced highest values using the NMS with respect to the cross sections of the information indication line taken along the lines perpendicular to the information indication line, in which values of adjacent pixels or hysterical thresholding employing different threshold values is used.

### Claim 11

A method of extracting a salient line for an information sign, the method comprising:
receiving an image including information indication lines to be detected;
generating a symmetric function signal consistent with a thickness and a direction of an information indication line having a preset thickness in the received image and performing a convolution of the information indication line and the generated symmetric function signal;
enhancing highest values in results of the convolution using a signal quality improvement filter; and
finding local highest values among the enhanced highest values and extracting a line connecting the local highest values as a salient line connecting center points in the information indication line.

### Claim 12 (depends on 11)

The method of claim 11, wherein the performing of the convolution includes generating a disk-shaped symmetric function signal consistent with an arbitrary information indication line having the preset thickness in all directions, and performing a convolution of the information indication line and the generated disk-shaped symmetric function signal.

### Claim 13 (depends on 11)

The method of claim 11, wherein the performing of the convolution includes generating a pillbox kernel which is a disk-shaped symmetric function defined as a function of image coordinates based on a center of the kernel and a radius of the kernel, and performing a convolution of the information indication line and the generated pillbox kernel.

### Claim 14 (depends on 13)

The method of claim 13, wherein the performing of the convolution includes approximating the generated pillbox kernel to a value obtained by dividing a sum of image data corresponding to a circular region of the pillbox kernel by an area of the circular region, and performing a convolution of the information indication line and the approximated pillbox kernel.

### Claim 15 (depends on 13)

The method of claim 13, wherein the performing of the convolution includes approximating a circular region of the generated pillbox kernel to a sum of a plurality of rectangular regions, and calculating the sum of the plurality of rectangular regions from sums of an integral image.

### Claim 16 (depends on 15)

The method of claim 15, wherein the performing of the convolution includes calculating the sum of the plurality of rectangular regions from the sums of the integral image using values at vertices in the integral image.

### Claim 17 (depends on 15)

The method of claim 15, wherein the performing of the convolution includes performing a convolution of the information indication line and the approximated pillbox kernel, and calculating a parameter minimizing an error function of areas of divided arcs and areas of divided rectangular regions using least-squares optimization.

### Claim 18 (depends on 11)

The method of claim 11, wherein the enhancing of the highest values includes enhancing the highest values of the convolution results using a morphological extrema filter (MEF) which is a signal quality improvement filter, and
the MEF is a sparse kernel having a plurality of pairs and is defined as a function of a filter gain and a preset threshold value.

### Claim 19 (depends on 11)

The method of claim 11, wherein the extracting of the salient line includes finding the local highest values in the enhanced highest values using non-maximum suppression (NMS) with respect to cross sections of the information indication line taken along lines perpendicular to the information indication line and extracting a line connecting the local highest values as the salient line connecting the center points in the information indication line.

### Claim 20 (depends on 19)

The method of claim 19, wherein the extracting of the salient line includes finding the local highest values in the enhanced highest values using the NMS with respect to the cross sections of the information indication line taken along the lines perpendicular to the information indication line, in which values of adjacent pixels or hysterical thresholding employing different threshold values is used.
