Apparatus and method for image processing to calculate likelihood of image of target object detected from input image

Abstract

An image processing apparatus, which receives an input image and detects an image of a target object based on a detection algorithm, includes a machine learning device which performs learning by using a plurality of partial images cut out from at least one input image, based on a result of detection of the image of the target object, and calculates a likelihood of the image of the target object.

What is claimed is: 
     
         1 . An image processing apparatus which receives an input image and detects an image of a target object based on a detection algorithm, comprising:
 a machine learning device which performs learning by using a plurality of partial images cut out from at least one input image, based on a result of detection of the image of the target object, and calculates a likelihood of the image of the target object.   
     
     
         2 . The image processing apparatus according to  claim 1 , wherein
 the machine learning device comprises:
 a state observation unit which observes, as a state variable, at least one of detection position, posture, size, and an image of a detected part related to the target object; and 
 a learning unit which generates a learning model for calculating the likelihood of the image of the target object, based on the state variable observed by the state observation unit, and 
   the machine learning device performs unsupervised learning.   
     
     
         3 . The image processing apparatus according to  claim 1 , wherein
 the machine learning device comprises:
 a state observation unit which observes, as a state variable, at least one of detection position, posture, size, and an image of a detected part related to the target object; 
 a label obtaining unit which obtains labels indicating correct detection or incorrect detection to be attached to each of the plurality of partial images; and 
 a learning unit which generates a learning model for calculating the likelihood of the image of the target object, based on the state variable observed by the state observation unit and the labels obtained by the label obtaining unit, and 
   the machine learning device performs supervised learning.   
     
     
         4 . The image processing apparatus according to  claim 3 , wherein
 the learning unit comprises:
 an error calculation unit which calculates errors based on the state variable observed by the state observation unit and the labels obtained by the label obtaining unit; and 
 a learning model update unit which updates the learning model based on outputs of the state observation unit and the error calculation unit. 
   
     
     
         5 . The image processing apparatus according to  claim 3 , wherein the learning unit attaches a label indicating incorrect detection to a partial image cut out from a region in the input image which region contains no image of the target object. 
     
     
         6 . The image processing apparatus according to  claim 1 , wherein the machine learning device uses an image obtained by converting an image contained in a predefined region with respect to a position and posture of the detected target object to calculate the likelihood of the image of the target object before or after performing the machine learning. 
     
     
         7 . The image processing apparatus according to  claim 1 , wherein the machine learning device receives features extracted by a same feature extraction method as used in the detection algorithm from an image contained in a predefined region with respect to a position and posture of the detected target object to calculate the likelihood of the image of the target object before or after performing the machine learning, and calculates the likelihood of the image of the target object. 
     
     
         8 . The image processing apparatus according to  claim 1 , wherein the machine learning device receives features extracted by a feature extraction method different from methods used in the detection algorithm from an image contained in a predefined region with respect to a position and posture of the detected target object to calculate the likelihood of the image of the target object before or after performing the machine learning. 
     
     
         9 . The image processing apparatus according to  claim 1 , wherein the machine learning device performs the machine learning step by step, allowing relatively large ranges for detection parameters at an initial stage and gradually allowing smaller ranges for the detection parameters as the learning progresses to subsequent stages. 
     
     
         10 . The image processing apparatus according to  claim 1 , wherein a threshold value against which to determine whether or not the image has been detected is established automatically based on the likelihood outputted by the machine learning device. 
     
     
         11 . The image processing apparatus according to  claim 1 , wherein the machine learning device is communicable with at least one other machine learning device and mutually exchange or share a learning model generated by the machine learning device with the at least one other machine learning device. 
     
     
         12 . An image processing method for receiving an input image and detecting an image of a target object based on a detection algorithm, comprising:
 performing machine learning by using a plurality of partial images cut out from at least one input image based on a detection result of the image of the target object to calculate a likelihood of the image of the target object.   
     
     
         13 . The image processing method according to  claim 12 ,
 wherein the performing of machine learning to calculate the likelihood of the image of the target object comprises:
 observing, as a state variable, at least one of detection position, posture, size, and an image of a detected part related to the target object; and 
 generating a learning model with which to calculate the likelihood of the image of the target object, based on the state variable, to perform unsupervised learning. 
   
     
     
         14 . The image processing method according to  claim 12 ,
 wherein the performing of machine learning to calculate the likelihood of the image of the target object comprises:
 observing, as a state variable, at least one of detection position, posture, size, and an image of a detected part related to the target object; 
 obtaining labels indicating correct detection or incorrect detection to be attached to each of the plurality of partial images; and 
 generating a learning model with which to calculate the likelihood of the image of the target object, based on the state variable and the labels, to perform supervised learning. 
   
     
     
         15 . The image processing method according to  claim 14 , wherein
 the generating of the learning model comprises:
 calculating errors based on the state variable and the labels; and 
 updating the learning model based on the state variable and the calculated errors.