System and method for evaluating a classifier implemented within an image signal processor

• US 9,842,280 B2
• Filed: 11/04/2015
• Issued: 12/12/2017
• Est. Priority Date: 11/04/2015
• Status: Active Grant

• Current Assignee: Omnivision Technologies Incorporated (Will Semiconductor Co., Ltd. Shanghai)
• Original Assignee: Omnivision Technologies Incorporated (Will Semiconductor Co., Ltd. Shanghai)
• Inventors: Hsu, Ming Kai
• Primary Examiner(s): MOYER, ANDREW M
• Application Number:
  US14/932,926
• Publication Number:
  US 20170124429A1
• Time in Patent Office: 769 Days
• Field of Search: 382159, 382160
• CPC Class Codes: G06F 18/2115 by evaluating different subsets according to an optimisation criterion, e.g. class separability, forward selection or backward elimination
• CPC Class Codes: G06F 18/2193 based on specific statistical tests
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
• CPC Class Codes: G06V 10/945 User interactive design; Environments; Toolboxes
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

A system for evaluating a classifier of an image signal processor (ISP) includes (i) a microprocessor and (ii) memory storing training images, the microprocessor being capable of sending each training image to the ISP. The system includes machine-readable instructions stored within the memory and executed by the microprocessor capable of: (i) selecting a subset of images based upon a divider position, (ii) controlling the ISP to classify each image as belonging or not belonging to an object class, (iii) determining a positive-match count, (iv) determining an error count based upon the positive-match count and total number of training images belonging to the object class, (v) repeating, for other divider positions, steps of selecting, controlling, and determining to identify an optimal divider position and a minimum-error count; and (vi) determining the classifier'"'"'s optimality by comparing the optimal divider position to a predetermined optimal divider position and a predetermined minimum-error count.

Claims

1. A system for evaluating a classifier implemented within an image signal processor (ISP) comprising:

a microprocessor and memory storing a training image set having a plurality of images, the microprocessor being capable of sending each of the plurality of images to the ISP, the plurality of images including Mtotal images indexable as 1, 2, . . . M; and

machine-readable instructions stored within the memory and executed by the microprocessor capable of;

selecting, based upon a divider position within the training image set denoted by a divider integer D between one and M+1 inclusive, the D^th image of the plurality of images;

controlling the ISP to classify the D^th image as belonging to or not belonging to an object class;

determining a positive-match count equal to zero when the D^th image is classified as not belonging to the object class and equal to one when the D^th image is classified as belonging to the object class;

determining an error count based upon (a) total number of images of the training image set belonging to the object class, and (b) the positive-match count;

repeating, for a plurality of other divider positions within the training image set, the steps of selecting, controlling, and determining to identify an optimal divider position corresponding to at least one of (a) a minimum-error count and (b) a maximum-error count, wherein total number classifier operations performed by the classifier to determine at least one of the minimum-error count and the maximum-error count equals total number of images in the training image set; and

determining optimality of the classifier by comparing at least one of (a) optimal divider position corresponding to a minimum-error count to a predetermined optimal divider position corresponding to a predetermined minimum-error count, and (b) optimal divider position corresponding to a maximum-error count to a predetermined optimal divider position and corresponding to a predetermined maximum-error count.

2. The system of claim 1, the machine-readable instructions being further capable of:

determining a negative-match count equal to one when the D^th image is classified as not belonging to the object class; and

determining the error count as proportional to the positive-match count subtracted from sum of (a) the negative-match count and (b) the total number of images of the training image set belonging to the object class.

3. The system of claim 1, the machine-readable instructions being further capable of:

determining, for the D^th image where D>

1, the error count as a cumulative error count partially based upon one previously-determined positive-match count and one previously-determined negative-match count for the (D−

1)^st image of the plurality of images.

4. The system of claim 1, wherein total number classifier operations performed by the classifier to determine at least one of the minimum-error count and the maximum-error count is a linear function of total number of images in the training image set.

5. The system of claim 1, the plurality of other divider positions being each other possible divider position.

6. A method for evaluating a classifier implemented within an image signal processor (ISP) to identify an object class in a received electronic image signal, the method comprising:

selecting, from a training image set having a plurality of images including M total images indexable as 1, 2, . . . , M, the D^th image of the plurality of images where D denotes a divider position within the training image set;

controlling the ISP to classify the D^th image as belonging to or not belonging to an object class;

determining a positive-match count equal to zero when the D^th image is classified as not belonging to the object class and equal to one when the D^th image is classified as belonging to the object class;

determining an error count based upon (a) total number of images of the training image set belonging to the object class, and (b) the positive-match count;

repeating, for a plurality of other divider positions within the training image set, the steps of selecting, controlling, and determining to identify an optimal divider position corresponding to at least one of (a) a minimum error count and (b) a maximum-error-count, wherein total number classifier operations performed by the classifier to determine at least one of the minimum-error count and the maximum-error count equals total number of images in the training image set; and

determining optimality of the classifier by comparing at least one of (a) optimal divider position corresponding to a minimum-error count to a predetermined optimal divider position corresponding to a predetermined minimum-error count, and (b) optimal divider position corresponding to a maximum-error count to a predetermined optimal divider position and corresponding to a predetermined maximum-error count.

7. The method of claim 6, further comprising:

determining a negative-match count equal to one when the D^th image is classified as not belonging to the object class; and

determining the error count as proportional to the positive-match count subtracted from sum of (a) the negative-match count and (b) the total number of images of the training image set belonging to the object class.

8. The method of claim 6,

the step of determining the error count further comprising, for the D^th image where D>

1, determining the error count as a cumulative error count partially based upon one previously-determined positive-match count and one previously-determined negative-match count for the (D−

1)^st image of the plurality of images.

9. The method of claim 6, the step of selecting further comprising selecting the D^th image such that total number classifier operations performed by the classifier to determine at least one of the minimum-error count and the maximum-error count is a linear function of total number of images in the training image set.

10. The method of claim 6, in the step of repeating, the plurality of other divider positions being each other possible divider position.