Apparatus and method for detecting object in image, and apparatus and method for computer-aided diagnosis

• US 10,255,673 B2
• Filed: 12/15/2015
• Issued: 04/09/2019
• Est. Priority Date: 12/15/2014
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Jeon, Joo Hyuk
• Primary Examiner(s): Wang, Carol
• Application Number:
  US14/969,108
• Publication Number:
  US 20160171341A1
• Time in Patent Office: 1,211 Days
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/75 involving models
• CPC Class Codes: G06V 10/443 by matching or filtering
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches

Abstract

An apparatus for detecting an object in an image, including a detection group determiner configured to classify a plurality of feature levels into one or more detection groups, wherein the plurality of feature levels includes multiple scales and are arranged as a feature pyramid; and a detection group processor configured to create one or more feature maps corresponding to the one or more detection groups, and to detect an object by applying one or more models to the one or more feature maps.

Claims

1. An apparatus for detecting an object in an image, the apparatus comprising:

a memory configured to store instructions; and

at least one processor, upon executing the stored instructions, configured to;

classify a plurality of feature levels included in the image into one a plurality of detection groups, wherein the plurality of feature levels includes multiple scales,

create a plurality of feature maps which respectively correspond to the plurality of detection groups from the image, wherein each of the plurality of feature maps includes a feature map which corresponds to a lowest feature level in each of the plurality of detection groups; and

detect an object by applying one or more models to the plurality of feature maps.

2. The apparatus of claim 1, wherein the at least one processor is further configured to:

generate a plurality of computational cost estimates corresponding to the plurality of feature levels, and

classify the plurality of feature levels into the plurality of detection groups using the plurality of computational cost estimates.

3. The apparatus of claim 2, wherein the plurality of computational cost estimates comprise:

a first plurality of feature generation computation costs associated with feature generation during application of a scale-invariant model,

a first plurality of score computation costs associated with score calculation during application of a scale-invariant model, and

a second plurality of score computation costs associated with score calculation during application of a multi-scale model.

4. The apparatus of claim 3, wherein the at least one processor is further configured to:

compare the first plurality of feature generation computation costs and the first plurality of score computation costs, to the second plurality of score computation costs, and

classify the plurality of feature levels into the plurality of detection groups based on the comparison.

5. The apparatus of claim 1, wherein each one of the plurality of feature maps is created at a scale corresponding to the lowest feature level in each one of the plurality of detection groups.

6. The apparatus of claim 1, wherein the at least one processor is further configured to determine a number of models to be applied to the plurality of detection groups based on one or more sizes associated with the plurality of detection groups.

7. The apparatus of claim 6, wherein the at least one processor is further configured to:

calculate one or more scores corresponding to the plurality of detection groups by applying a plurality of scaled filters to the plurality of feature maps, wherein a number of the plurality of scaled filters corresponds to the number of models, and

detect the object using the one or more scores.

8. The apparatus of claim 7, wherein the at least one processor is further configured to:

adjust a frequency of detection based on a size of the object to be detected, and

calculate the one or more scores based on the adjusted frequency of detection.

9. A method for detecting an object in an image, comprising:

classifying a plurality of feature levels included in the image into a plurality of detection groups, wherein the plurality of feature levels includes multiple scales;

creating a plurality of feature maps which respectively correspond to the plurality of detection groups, wherein each of the plurality of feature maps includes a feature map which corresponds to a lowest feature level in each of the plurality of detection groups; and

detecting an object by applying one or more models to the plurality of feature maps.

10. The method of claim 9, wherein the classifying the plurality of feature levels into the plurality of detection groups comprises:

generating a plurality of computational cost estimates associated with the plurality of feature levels; and

classifying the plurality of feature levels into the plurality of detection groups using the plurality of computational cost estimates.

11. The method of claim 10, wherein the plurality of computational cost estimates include:

a first plurality of feature generation computation costs associated with feature generation during application of a scale-invariant model;

a first plurality of score computation costs associated with score calculation during application of a scale-invariant model; and

a second plurality of score computation costs associated with score calculation during application of a multi-scale model.

12. The method of claim 11, wherein the classifying the plurality of feature levels into the plurality of detection groups includes:

comparing the first plurality of feature generation computation costs and the first plurality of score computation costs, with the second plurality of score computation costs; and

classifying the plurality of feature levels into the plurality of detection groups based on the comparison.

13. The method of claim 11, wherein the classifying the plurality of feature levels into the plurality of detection groups comprises:

creating a first detection group that includes a lowest feature level of the plurality of feature levels;

calculating a first computation cost by combining a first feature generation computation cost of the first plurality of feature generation computation costs with a first score computation costs of the first plurality of score computation costs;

comparing the first computation cost with a second score computation cost of the second plurality of score computation costs associated with a first feature level that is higher than a highest feature level of the first detection group;

in response to the first computation cost being greater than the second score computation cost, adding the first feature level to the first detection group; and

in response to the first computation cost being less than or equal to than the second score computation cost, creating a second detection group to include the first feature level.

14. The method of claim 9, wherein the creating of the plurality of feature maps includes creating a feature map at a level that corresponds to the lowest feature level in each one of the plurality of detection groups.

15. The method of claim 9, wherein the detection of the object comprises determining a number of models to be applied to the plurality of detection groups based on one or more sizes associated with the plurality of detection groups.

16. The method of claim 15, wherein the detecting the object comprises:

calculating one or more scores corresponding to the plurality of detection groups by applying a plurality of scaled filters to the plurality of feature maps, wherein a number of the plurality of scaled filters corresponds to the number of models; and

detecting the object using the one or more scores.

17. The method of claim 16, wherein the detecting the object further comprises adjusting a frequency of detection based on a size of the object to be detected, and

wherein the calculating the one or more scores further comprises calculating the one or more scores based on the adjusted frequency of detection.

18. A computer-aided diagnosis (CAD) apparatus comprising:

a memory configured to store instructions; and

at least one processor, upon executing the stored instructions, configured to;

in response to receiving a medical image, classify a scale-space of the medical image into a plurality of detection groups,

create a plurality of feature maps which respectively correspond to the plurality of detection groups from the received medical image, wherein each of the plurality of feature maps includes a feature map which corresponds to a lowest feature level in each of the plurality of detection groups,

detect a lesion by applying one or more models to the plurality of feature maps, and

determine whether the detected lesion is benign.

19. The CAD apparatus of claim 18, wherein, with respect to a plurality of scales in the scale-space, the at least one processor is further configured to:

estimate a first plurality of feature generation computation costs and a first plurality of score computation costs associated with applying a scale-invariant model,

estimate a second plurality of score computation costs by associated with applying a multi-scale model, and

classify the plurality of scales into the plurality of detection groups by taking into consideration the first plurality of feature generation computation costs, the first plurality of score computation costs, and the second plurality of score computation costs.

20. The CAD apparatus of claim 18, wherein the at least one processor is further configured to:

determine a number of the one or more models based on one or more sizes associated with the plurality of detection groups,

calculate a plurality of scores associated with a plurality of locations by applying one or more filters of different scales, wherein a number of the one or more filters is equal to the number of the one or more models, and

detect that a location of the plurality of locations whose score estimate is greater than a designated threshold is the lesion.