1. Automated thumbnail selection for online video
Inventors: Sophie Lebrecht, Michael J. Tarr, Deborah Johnson, Mark Desnoyer, Sunil Mallya Kasaragod
Summary: Access is provided to optimal thumbnails that are extracted from a stream of video. Using a processing device configured with a deep learning model that incorporates preferences generated by the brain and behavior from the perception of visual images, the optimal thumbnail(s) for a given video is/are selected, stored and/or displayed.
2. Selecting a high-valence representative image
Inventors: Mark Desnoyer, David Lea, Sophie Lebrecht, Padraig Michael Furlong, Nicholas Dufour, Sunil Mallya, Deborah Johnson, Michael J. Tarr
Summary: A collection of images, such as frames from a video clip, movie, image album, live video feed, etc., are received. A user requests a thumbnail picture representative of the plurality of images. The plurality of images is filtered by various means to obtain a set of images. The filtering can be based on a blurriness of the image, whether an image is near a scene transition, an amount of text depicted in the image, or a color level of the image. Valence scores, which can predict user responses to images, may be determined for one or more of the images in the remaining set of images. A first image from the set of images is selected based at least in part on the valence score of the first image. The first image is sent for display. Images are subsequently tested using an iterative A/B testing setup.
3. Method and system for using neuroscience to predict consumer preference
Inventors: Sophie Lebrecht, Michael J. Tarr, David Sheinberg, Moshe Bar
Summary: A system and method provides techniques and analysis tools for measuring how individuals perceive and respond to valences and more subtle "micro-valences" present in stimuli. The invention includes a process that uses human neuroimaging and behavioral techniques to measure valence in order to predict how individuals will perceive and react to any stimulus designed to engage individuals, e.g., end users or consumers, including, but not limited to, products, brands, logos, packaging, banner ads, and advertisements and their subcomponents or features, such as shape, color, pattern, and material properties.
4. Method for clustering novel facial images based on identity
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour, Michael Rossiter, Mridul Khan, David Lea
Summary: This patent describes the use of deep convolutional neural networks to group faces by identity, invariant of image type or content orientation. The methods outlined in this patent can also be applied to objects.
Download patent
5. A method for measuring implicit visual responses with an online task
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent involves modifying Neon’s proprietary “Birthday Task” and using the scores from the task to create a ranking scale of image preference. This ranking scale would become our “gold standard” dataset, which would be used to build our models.
Download patent
6. Selecting a personalized high-valence representative image
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent outlines methods for personalizing images for different users, platforms, and devices. It additionally describes: how Neon constructs clusters of people that share a similar image preference perception; and how Neon can translate that information based on more standard metadata on individual users that a large scale platform would have.
Download patent
7. Selecting a high-valence representative image based on image quality
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: By understanding the relationship between image resolution, valence, and CTR, we can recommend images to present to users in low bandwidth environments. The application of these techniques is of value for CDNs that are serving images in a variety of bandwidths. This is also important for companies that are serving images on a variety of screen sizes and at a variety of resolutions.
Download patent
8. Automatic optimization of image capture on mobile devices by human and non-human agents
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour, Zhihao Li, Nicole Halmi, David Sheinberg, Michael J. Tarr
Summary: We present methods for the programmatic and personalized capture and enhancement of high-valence images and videos in real-time, achieved by running image selection software on a mobile phone or other digital camera to control image selection at the point of capture, as opposed to filtering and selecting images and video that have already been captured.
Download patent
9. Method for determining image similarity in a multidimensional valence feature space
Inventors: Mark Desnoyer, Sophie Lebrecht, Zhihao Li, Nicholas Dufour
Summary: This patent describes methods to group images by style using valence and the valence feature space. These methods can be employed to, for example, improve search recommendations, make cross-category image recommendations, identify the tightness of a cluster of images in terms of unified style, and identify users with similar styles.
Download patent
10. Method for measuring implicit valence of images through web events
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent describes methods for building models that allow for the determination of image valence based on click data from the wild, implicitly, without any experimental manipulation. This method allows for a “closed” loop between a controlled experimental setup (e.g. the task described in patent 5 above) and click data obtained about images that are live on websites. This approach allows for the exponential and effortless scaling of data in deep learning models by simply serving and tracking the CTR of an image.
Download patent
11. Systems and methods of generating a modified-view video from a video file based on
valence
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent outlines methods to automatically crop video in a way that maximizes valence, action, or other desired properties while maintaining visual continuity. This method can be used to automatically transform horizontal video into vertical video, and can be combined with image filters and personalization. This method is useful as advertisers start wanting to adapt their existing horizontal/traditional video content to fit mobile-first vertical video and social platforms.
Download patent
Inventors: Sophie Lebrecht, Michael J. Tarr, Deborah Johnson, Mark Desnoyer, Sunil Mallya Kasaragod
Summary: Access is provided to optimal thumbnails that are extracted from a stream of video. Using a processing device configured with a deep learning model that incorporates preferences generated by the brain and behavior from the perception of visual images, the optimal thumbnail(s) for a given video is/are selected, stored and/or displayed.
2. Selecting a high-valence representative image
Inventors: Mark Desnoyer, David Lea, Sophie Lebrecht, Padraig Michael Furlong, Nicholas Dufour, Sunil Mallya, Deborah Johnson, Michael J. Tarr
Summary: A collection of images, such as frames from a video clip, movie, image album, live video feed, etc., are received. A user requests a thumbnail picture representative of the plurality of images. The plurality of images is filtered by various means to obtain a set of images. The filtering can be based on a blurriness of the image, whether an image is near a scene transition, an amount of text depicted in the image, or a color level of the image. Valence scores, which can predict user responses to images, may be determined for one or more of the images in the remaining set of images. A first image from the set of images is selected based at least in part on the valence score of the first image. The first image is sent for display. Images are subsequently tested using an iterative A/B testing setup.
3. Method and system for using neuroscience to predict consumer preference
Inventors: Sophie Lebrecht, Michael J. Tarr, David Sheinberg, Moshe Bar
Summary: A system and method provides techniques and analysis tools for measuring how individuals perceive and respond to valences and more subtle "micro-valences" present in stimuli. The invention includes a process that uses human neuroimaging and behavioral techniques to measure valence in order to predict how individuals will perceive and react to any stimulus designed to engage individuals, e.g., end users or consumers, including, but not limited to, products, brands, logos, packaging, banner ads, and advertisements and their subcomponents or features, such as shape, color, pattern, and material properties.
4. Method for clustering novel facial images based on identity
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour, Michael Rossiter, Mridul Khan, David Lea
Summary: This patent describes the use of deep convolutional neural networks to group faces by identity, invariant of image type or content orientation. The methods outlined in this patent can also be applied to objects.
Download patent
5. A method for measuring implicit visual responses with an online task
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent involves modifying Neon’s proprietary “Birthday Task” and using the scores from the task to create a ranking scale of image preference. This ranking scale would become our “gold standard” dataset, which would be used to build our models.
Download patent
6. Selecting a personalized high-valence representative image
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent outlines methods for personalizing images for different users, platforms, and devices. It additionally describes: how Neon constructs clusters of people that share a similar image preference perception; and how Neon can translate that information based on more standard metadata on individual users that a large scale platform would have.
Download patent
7. Selecting a high-valence representative image based on image quality
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: By understanding the relationship between image resolution, valence, and CTR, we can recommend images to present to users in low bandwidth environments. The application of these techniques is of value for CDNs that are serving images in a variety of bandwidths. This is also important for companies that are serving images on a variety of screen sizes and at a variety of resolutions.
Download patent
8. Automatic optimization of image capture on mobile devices by human and non-human agents
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour, Zhihao Li, Nicole Halmi, David Sheinberg, Michael J. Tarr
Summary: We present methods for the programmatic and personalized capture and enhancement of high-valence images and videos in real-time, achieved by running image selection software on a mobile phone or other digital camera to control image selection at the point of capture, as opposed to filtering and selecting images and video that have already been captured.
Download patent
9. Method for determining image similarity in a multidimensional valence feature space
Inventors: Mark Desnoyer, Sophie Lebrecht, Zhihao Li, Nicholas Dufour
Summary: This patent describes methods to group images by style using valence and the valence feature space. These methods can be employed to, for example, improve search recommendations, make cross-category image recommendations, identify the tightness of a cluster of images in terms of unified style, and identify users with similar styles.
Download patent
10. Method for measuring implicit valence of images through web events
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent describes methods for building models that allow for the determination of image valence based on click data from the wild, implicitly, without any experimental manipulation. This method allows for a “closed” loop between a controlled experimental setup (e.g. the task described in patent 5 above) and click data obtained about images that are live on websites. This approach allows for the exponential and effortless scaling of data in deep learning models by simply serving and tracking the CTR of an image.
Download patent
11. Systems and methods of generating a modified-view video from a video file based on
valence
Inventors: Mark Desnoyer, Sophie Lebrecht, Nicholas Dufour
Summary: This patent outlines methods to automatically crop video in a way that maximizes valence, action, or other desired properties while maintaining visual continuity. This method can be used to automatically transform horizontal video into vertical video, and can be combined with image filters and personalization. This method is useful as advertisers start wanting to adapt their existing horizontal/traditional video content to fit mobile-first vertical video and social platforms.
Download patent