System and method for image quality scoring
A system receives images of objects. The system identifies a category for each of the objects, and extracts features from the images. The features relate to a quality of the image. The features of the images are stored in a database according to the category of each object, such that each set of features is associated with its corresponding image. The system displays the images on a network-based publication system, and receives data relating to the displayed images. The data is analyzed, and the images are ranked as a function of the analysis. The system redisplays the images on the network-based publication system as a function of the ranking of the images.
1. A system comprising:
one or more processors;
a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
accessing images of objects;
segmenting foregrounds of the images to identify the objects depicted in the segmented foregrounds of the images;
determining a respective multidimensional feature vector for the each of the respective images, wherein determining each multidimensional feature vector includes determining a vector for each of a plurality of features comprising a respective size, a respective brightness, and a respective contrast for each of the respective objects within the respective images based on the segmented foregrounds of the images;
receiving data that indicates at least some of the objects depicted in the segmented foregrounds of the images were purchased by users during display of the images on a network-based publication system;
for each of the images, calculating a ratio of purchases of the object depicted in the image to presentations of the image by the network-based publication system;
for each of the images, correlating, by a machine learning phase each feature vector of the multidimensional feature vector for the image to the calculated ratio of purchase to presentation for the corresponding image, and selecting a feature vector of the multidimensional feature vector that is most closely related to the calculated ratio of purchases to presentations for the image;
ranking the images based on a combination of the calculated ratios of purchases to presentations and the selected feature vector; and
redisplaying the images on the network-based publication system to at least one user based on the ranking of the images.
2. The system of claim 1 , wherein the ranking of the images comprises generating a model for the images.
3. The system of claim 2 , wherein the ranking of the images is a first ranking of images and is further based on a factor selected from a group consisting of a previous second ranking of the images, a price at which at least one of the objects was sold, a price-sales function, a sales-time on site function, and a human scored quality and price function.
4. The system of claim 3 , wherein the ranking of the images is further based on data that indicates a user action selected from a group consisting of selecting an object by clicking on the object, placing the object in an electronic shopping cart, and viewing the object for a measured amount of time.
5. The system of claim 1 , wherein the network-based publication system comprises an electronic auction system.
6. The system of claim 1 , wherein the objects comprise items offered for sale on the network-based publication system.
7. The system of claim 1 , wherein the ranking of the image is further based on a feature selected from a group consisting of a contrast between the object and a background of the image, a clarity of the image, and a color spectrum of the image.
8. The system of claim 1 , wherein the redisplaying of the images on the network-based publication system comprises displaying, on a search results page, images with a higher ranking above images with a lower ranking.
9. A method comprising:
by operation of one or more processors of a machine:
accessing images of objects;
segmenting foregrounds of the images to identify the objects depicted in the segmented foregrounds of the images;
determining a respective multidimensional feature vector for the each of the respective images, wherein determining each multidimensional feature vector includes determining a vector for each of a plurality of features comprising a respective size, a respective brightness, and a respective contrast for each of the respective objects within the respective images based on the segmented foregrounds of the images;
receiving data that indicates at least some of the object depicted in the segmented foregrounds of the images were purchased by users during display of the images on a network-based publication system;
for each of the images, calculating a ratio of purchases of the object depicted in the image to presentations of the image by the network-based publication system;
for each of the images, correlating, by a machine learning phase each feature vector of the multidimensional feature vector for the image to the calculated ratio of purchase to presentation for the corresponding image, and selecting a feature vector of the multidimensional feature vector that is most closely related to the calculated ratio of purchases to presentations for the image;
ranking the images based on a combination of the calculated ratios of purchases to presentations and the selected feature vector; and
redisplaying the images on the network-based publication system to at least one user based on the ranking of the images.
10. The method of claim 9 ,
wherein the ranking of the images comprises generating a model for the images;
wherein the ranking of the images is a first ranking of images and is further based on a factor selected from a group consisting of a previous second ranking of the images, a price at which the object sold, a price-sales function, a sales-time on site function, and a human scored quality and price function; and
wherein the ranking of the image is further based on data that indicates a user action selected from a group consisting of selecting an object by clicking on the object, placing the object in an electronic shopping cart, and viewing the object for a measured amount of time.
11. The method of claim 9 , wherein the ranking of the images is further based on a feature selected from a group consisting of a contrast between the object and a background of the image, a clarity of the image, and a color spectrum of the image.
12. The method of claim 9 , wherein the redisplaying of the images on the network-based publication system comprises displaying, on a search results page, images with a higher ranking above images with a lower ranking.
13. A non-transitory computer readable storage device comprising instructions that, when executed by one or more processors of a machine, cause the machine to perform operations comprising:
accessing images of objects;
determining a respective multidimensional feature vector for the each of the respective images, wherein determining each multidimensional feature vector includes determining a vector for each of a plurality of features comprising a respective size, a respective brightness, and a respective contrast for each of the respective objects within the respective images based on the segmented foregrounds of the images;
receiving data that indicates at least some of the object depicted in the segmented foregrounds of the images were purchased by users during display of the images on a network-based publication system;
for each of the images, calculating a ratio of purchases of the object depicted in the image to presentations of the image by the network-based publication system;
for each of the images, correlating, by a machine learning phase each feature vector of the multidimensional feature vector for the image to the calculated ratio of purchase to presentation for the corresponding image, and selecting a feature vector of the multidimensional feature vector that is most closely related to the calculated ratio of purchases to presentations for the image;
ranking the images based on a combination of the calculated ratios of purchases to presentations and the selected feature vector; and
redisplaying the images on the network-based publication system to at least one user based on the ranking of the images.
14. The non-transitory computer readable storage device of claim 13 ,
wherein the ranking of the images comprises generating a model for the images;
wherein the ranking of the images is a first ranking of images and is further based on a factor selected from a group consisting of a previous second ranking of the images, a price at which the object sold, a price-sales function, a sales-time on site function, and a human scored quality and price function; and
wherein the ranking of the image is further based on data that indicates a user action selected from a group consisting of selecting an object by clicking on the object, placing the object in an electronic shopping cart, and viewing the object for a measured amount of time.
15. The non-transitory computer readable storage device of claim 13 , wherein the redisplaying of the images on the network-based publication system comprises displaying, on a search results page, images with a higher ranking above images with a lower ranking.