Ordering image search results

Abstract

Methods, systems, and apparatus, including computer program products, for ranking images are disclosed. An image search subsystem generates an adjustment factor representative of a quality measure of an image relative to a search query. The quality represents a relevance of the image to the query. The adjustment factor can be computed based on relevance data for the image to the query and image similarity data representing a relative similarity between the image and other images relevant to the query. The relevance data can be based on user actions in response to the image being included in search results for the query. The adjustment factor can be scaled based on whether the relevance data and the image similarity data both indicate that the image is relevant to the search query. A relevance score is computed based on the adjustment factor (e.g., a product of the adjustment factor and relevance score).

1 - 20 . (canceled) 
     
     
         21 . A computer-implemented method comprising:
 identifying, independent of visual similarity between images, a particular image that has a measure of relevance to a specified search query that meets a specified relevance threshold;   determining that the particular image has at least a threshold level of visual similarity to one or more other images that are considered relevant to the specified search query;   determining, for the particular image, an adjustment factor that is based at least in part on the measure of relevance to the specified search query and the visual similarity level; and   determining, for the particular image, an adjusted measure of relevance to the specified search query in response to the measure of relevance meeting the specified relevance threshold and the particular image having at least the threshold level of visual similarity to the one or more other images, the adjusted measure of relevance being proportionate to the measure of relevance and the adjustment factor.   
     
     
         22 . The method of  claim 21 , further comprising:
 receiving a request for image search results responsive to the specified search query;   identifying a set of image search results that are responsive to the specified query, the set of image search results including a particular image search result referencing the particular image; and   ordering the set of image search results based at least in part on the adjusted measure of relevance for the particular image.   
     
     
         23 . The method of  claim 21 , wherein the specified relevance threshold is a minimum relevance score required for a measure of relevance for an image to be eligible to be positively adjusted, and the threshold level of visual similarity is a minimum level of visual similarity required for a measure of relevance for an image to be eligible to be positively adjusted. 
     
     
         24 . The method of  claim 21 , wherein adjusting the measure of relevance comprises determining a product of an amplification factor and the measure of relevance to the specified search query. 
     
     
         25 . The method of  claim 24 , further comprising:
 obtaining an image similarity score for the particular image, the image similarity score indicating a measure of visual similarity between the particular image and the one or more other images that are considered relevant to the specified search query; and   determining the amplification factor based on a product of the image similarity score for the particular image and the measure of relevance for the particular image.   
     
     
         26 . The method of  claim 21 , wherein the determination that the particular image has at least the threshold level of visual similarity to the one or more other images that are considered relevant to the specified search query is based on a visual distance matrix that specifies similarities between the particular image and the one or more other images, the similarities being specified by visual distance measures between image features of the particular image and image features of the one or more other images. 
     
     
         27 . The method of  claim 21 , further comprising:
 identifying, independent of visual similarity between images, a second image that has a second measure of relevance to the specified search query that meets a specified irrelevance threshold;   determining that the second image has at least a threshold level of visual dissimilarity to the one or more other images that are considered relevant to the specified search query; and   adjusting, for the second image, the second measure of relevance to the specified search query in response to the second measure of relevance meeting the specified irrelevance threshold and the second image having at least the threshold level of visual dissimilarity to the one or more other images.   
     
     
         28 . The method of  claim 27 , wherein the specified irrelevance threshold is a maximum relevance score allowed for a measure of relevance for an image to be eligible to be negatively adjusted, and the threshold level of visual dissimilarity is a maximum level of visual similarity allowed for a measure of relevance for an image to be eligible to be negatively adjusted. 
     
     
         29 . A system, comprising:
 a data processing apparatus; and   a data store storing instructions that, when executed by the data processing apparatus, cause the data processing apparatus to perform operations comprising:   identifying, independent of visual similarity between images, a particular image that has a measure of relevance to a specified search query that meets a specified relevance threshold;   determining that the particular image has at least a threshold level of visual similarity to one or more other images that are considered relevant to the specified search query;   determining, for the particular image, an adjustment factor that is based at least in part on the measure of relevance to the specified search query and the visual similarity level; and   determining, for the particular image, an adjusted measure of relevance to the specified search query in response to the measure of relevance meeting the specified relevance threshold and the particular image having at least the threshold level of visual similarity to the one or more other images, the adjusted measure of relevance being proportionate to the measure of relevance and the adjustment factor.   
     
     
         30 . The system of  claim 29 , wherein the operations further comprise: receiving a request for image search results responsive to the specified search query;
 identifying a set of image search results that are responsive to the specified query, the set of image search results including a particular image search result referencing the particular image; and   ordering the set of image search results based at least in part on the adjusted measure of relevance for the particular image.   
     
     
         31 . The system of  claim 29 , wherein the specified relevance threshold is a minimum relevance score required for a measure of relevance for an image to be eligible to be positively adjusted, and the threshold level of visual similarity is a minimum level of visual similarity required for a measure of relevance for an image to be eligible to be positively adjusted. 
     
     
         32 . The system of  claim 29 , wherein adjusting the measure of relevance comprises determining a product of an amplification factor and the measure of relevance to the specified search query. 
     
     
         33 . The system of  claim 32 , wherein the operations further comprise:
 obtaining an image similarity score for the particular image, the image similarity score indicating a measure of visual similarity between the particular image and the one or more other images that are considered relevant to the specified search query; and   determining the amplification factor based on a product of the image similarity score for the particular image and the relevance score for the particular image.   
     
     
         34 . The system of  claim 29 , wherein the determination that the particular image has at least the threshold level of visual similarity to the one or more other images that are considered relevant to the specified search query is based on a visual distance matrix that specifies similarities between the particular image and the one or more other images, the similarities being specified by visual distance measures between image features of the particular image and image features of the one or more other images. 
     
     
         35 . The system of  claim 29 , wherein the operations further comprise:
 identifying, independent of visual similarity between images, a second image that has a second measure of relevance to the specified search query that meets a specified irrelevance threshold;   determining that the second image has at least a threshold level of visual dissimilarity to the one or more other images that are considered relevant to the specified search query; and   adjusting, for the second image, the second measure of relevance to the specified search query in response to the second measure of relevance meeting the specified irrelevance threshold and the second image having at least the threshold level of visual dissimilarity to the one or more other images.   
     
     
         36 . The system of  claim 35 , wherein the specified irrelevance threshold is a maximum relevance score allowed for a measure of relevance for an image to be eligible to be negatively adjusted, and the threshold level of visual dissimilarity is a maximum level of visual similarity allowed for a measure of relevance for an image to be eligible to be negatively adjusted. 
     
     
         37 . A computer readable medium encoded with a computer program comprising instructions that when executed operate to cause a computer to perform operations comprising:
 identifying, independent of visual similarity between images, a particular image that has a measure of relevance to a specified search query that meets a specified relevance threshold;   determining that the particular image has at least a threshold level of visual similarity to one or more other images that are considered relevant to the specified search query;   determining, for the particular image, an adjustment factor that is based at least in part on the measure of relevance to the specified search query and the visual similarity level; and   determining, for the particular image, an adjusted measure of relevance to the specified search query in response to the measure of relevance meeting the specified relevance threshold and the particular image having at least the threshold level of visual similarity to the one or more other images, the adjusted measure of relevance being proportionate to the measure of relevance and the adjustment factor.   
     
     
         38 . The computer readable medium of  claim 37 , wherein the operations further comprise:
 receiving a request for image search results responsive to the specified search query;   identifying a set of image search results that are responsive to the specified query, the set of image search results including a particular image search result referencing the particular image; and   ordering the set of image search results based at least in part on the adjusted measure of relevance for the particular image.   
     
     
         39 . The computer readable medium of  claim 37 , wherein the specified relevance threshold is a minimum relevance score required for a measure of relevance for an image to be eligible to be positively adjusted, and the threshold level of visual similarity is a minimum level of visual similarity required for a measure of relevance for an image to be eligible to be positively adjusted. 
     
     
         40 . The computer readable medium of  claim 37 , wherein adjusting the measure of relevance comprises determining a product of an amplification factor and the measure of relevance to the specified search query.