System and method for recognizing offensive images

Abstract

According to one aspect, a method for categorizing at least one image includes obtaining the at least one image and mapping the at least one image to at least a first grid. The first grid is a two-dimensional grid that includes a plurality of cells. The method also includes characterizing the first grid, wherein categorizing the first grid includes determining whether the first grid is indicative of an offensive characteristic, and identifying the at least one image as offensive when it is determined that the first grid is indicative of the offensive characteristic. When it is determined that the first grid is not indicative of the offensive characteristic, the at least one image is identified as not offensive.

1 .- 25 . (canceled) 
     
     
         26 . A method, comprising:
 maintaining, by a device, a model of an offensive human pose, wherein the model indicates locations of skin-colored blocks in a first grid;   dividing, by the device, an image into a second grid having a plurality of blocks, each block of the second grid comprising one or more pixels of the image;   identifying, by the device, skin-colored blocks in the second grid;   making, by the device, a comparison between locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid; and   determining, by the device, that the image depicts the offense human pose based on the comparison between the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid.   
     
     
         27 . The method as in  claim 26 , wherein identifying skin-colored blocks in the second grid comprises:
 determining, by the device, whether a particular block in the second grid comprises a threshold number of skin-colored pixels.   
     
     
         28 . The method as in  claim 27 , wherein determining whether the particular block in the second grid comprises a threshold number of skin-colored pixels comprises:
 determining, by the device, a ratio of skin-colored pixels in the particular block to pixels in the particular block that are not skin-colored.   
     
     
         29 . The method as in  claim 27 , wherein determining whether the particular block in the second grid comprises a threshold number of skin-colored pixels comprises:
 determining, by the device, probabilities of each pixel in the particular block being classified as skin-colored.   
     
     
         30 . The method as in  claim 27 , wherein determining whether the particular block in the second grid comprises a threshold number of skin-colored pixels comprises:
 assessing, by the device, a connection between adjacent skin-colored pixels in the particular block.   
     
     
         31 . The method as in  claim 26 , wherein making the comparison between the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid comprises:
 using a machine-learning classifier to compare the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid.   
     
     
         32 . The method as in  claim 26 , further comprising:
 determining, by the device, a classification for a webpage or email that includes the image.   
     
     
         33 . The method as in  claim 26 , wherein the device determines that the image depicts the offense human pose based on a plurality of comparisons between the locations of the skin-colored blocks in the second grid to locations of skin-colored blocks in a plurality of models of different offensive human poses. 
     
     
         34 . The method as in  claim 26 , further comprising:
 associating, by the device, a weight to the determination that the image depicts the offensive human pose based on a measure of quality of the image.   
     
     
         35 . An apparatus, comprising:
 a processor; and   a memory configured to store a process executable by the processor, the process when executed by the processor configured to:
 maintain a model of an offensive human pose, wherein the model indicates locations of skin-colored blocks in a first grid; 
 divide an image into a second grid having a plurality of blocks, each block of the second grid comprising one or more pixels of the image; 
 identify skin-colored blocks in the second grid; 
 make a comparison between locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid; and 
 determine that the image depicts the offense human pose based on the comparison between the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid. 
   
     
     
         36 . The apparatus as in  claim 35 , wherein the apparatus identifies the skin-colored blocks in the second grid by:
 determining whether a particular block in the second grid comprises a threshold number of skin-colored pixels.   
     
     
         37 . The apparatus as in  claim 36 , wherein the apparatus determines whether the particular block in the second grid comprises a threshold number of skin-colored pixels by:
 determining a ratio of skin-colored pixels in the particular block to pixels in the particular block that are not skin-colored.   
     
     
         38 . The apparatus as in  claim 36 , wherein the apparatus determines whether the particular block in the second grid comprises a threshold number of skin-colored pixels by:
 determining probabilities of each pixel in the particular block being classified as skin-colored.   
     
     
         39 . The apparatus as in  claim 36 , wherein the apparatus determines whether the particular block in the second grid comprises a threshold number of skin-colored pixels by:
 assessing a connection between adjacent skin-colored pixels in the particular block.   
     
     
         40 . The apparatus as in  claim 35 , wherein the apparatus makes the comparison between the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid by:
 using a machine-learning classifier to compare the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid.   
     
     
         41 . The apparatus as in  claim 35 , wherein the process when executed is further configured to:
 determine a classification for a webpage or email that includes the image.   
     
     
         42 . The apparatus as in  claim 35 , wherein the apparatus determines that the image depicts the offense human pose based on a plurality of comparisons between the locations of the skin-colored blocks in the second grid to locations of skin-colored blocks in a plurality of models of different offensive human poses. 
     
     
         43 . The apparatus as in  claim 35 , wherein the process when executed is further configured to:
 associate a weight to the determination that the image depicts the offensive human pose based on a measure of quality of the image.   
     
     
         44 . A non-transitory computer-readable medium comprising instructions that, when executed by a processor, cause the processor to:
 maintain a model of an offensive human pose, wherein the model indicates locations of skin-colored blocks in a first grid;   divide an image into a second grid having a plurality of blocks, each block of the second grid comprising one or more pixels of the image;   identify skin-colored blocks in the second grid;   make a comparison between locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid; and   determine that the image depicts the offense human pose based on the comparison between the locations of the skin-colored blocks in the second grid and the locations of the skin-colored blocks in the first grid.   
     
     
         45 . The non-transitory computer-readable medium as in  claim 44 , wherein the model of the offensive human pose comprises a two-dimensional Boolean grid.