杏吧原创

Customise your favourite TV show

Plot-deconstructing software lets you make your own episodes of favourite shows
Need a reminder of his story?
Need a reminder of his story?
(Image: HBO/Everett/Rex Features)

LOST the plot watching Homeland or Game of Thrones? Wondering when a strange character you鈥檇 never seen before on Doctor Who was introduced? You鈥檙e not alone: the tremendous choice of programmes on offer on today鈥檚 multichannel TV services can make it hard to keep up.

But help is at hand, thanks to scene-analysis software that can compile a video sequence summarising any chosen plot line or character鈥檚 appearances in a TV series. Choose a scene, for instance, and the software will assemble a personalised video episode based around it. And in a move screenwriters will doubtless detest, it can also help fans compile customised episodes starring only their favourite actors.

聯If you are interested in a certain actor, the software will compile a summary of scenes they appear in聰

Called StoryVisualizer 鈥 or StoViz 鈥 the has been made possible by the increasing amount of content that can be stored on digital video recorders, says lead developer Herv茅 Bredin of the Computer Sciences Laboratory for Mechanics and Engineering Sciences (LIMSI) in Orsay, France. TiVo and Sky Plus, for example, can store up to a week of continuous broadcast TV plus all of the previous episodes in various series. 鈥淲e think it鈥檚 time users had search and browsing tools to use within these digital video collections,鈥 Bredin says.

He worked with a team at the to make that possible, exploiting the brawn of today鈥檚 graphics processing chips.

The task required more than just good hardware, though. Identifying key plot points and tracing character arcs are not easy, even for a human. In fact what makes a show compelling is often the way it weaves these elements together in sophisticated ways.

So StoViz鈥檚 first task involved 鈥渄e-interlacing鈥 those themes into individual threads. To find a particular storyline, it used image analysis to seek actors鈥 faces and background scenery used in certain scenarios, in addition to analysing the audio for key words associated with that story.

The software then assembles a group of scenes that its video and audio algorithms have decided are semantically similar, and therefore hopefully represents the same plot line. In the same way, if you are interested in a certain actor, you can choose their face and only their scenes will be compiled into a summary.

The researchers successfully tested StoViz on three TV series with very different formats: the sitcom Malcolm in the Middle, the fantasy drama Game of Thrones and the legal lunacy of Ally McBeal.

Now all that remains is to integrate the technology into existing entertainment systems. There are a number of possibilities. The system could easily be activated via a screen menu, says Bredin, and perhaps in the not-too-distant future you might simply say the name of your favourite character to a Siri-like voice interface (see 鈥Coming soon鈥 talking to your TV鈥).

Marek Barwinski, a vision engineer with imaging-app maker Cortexica in London, says that the 鈥渇airly restricted number of characters鈥 in most TV series means the processing won鈥檛 be too demanding. 鈥淏ut I wonder if in large video collections it could keep up with soap characters as they age,鈥 he says.

The device will be useful in some situations, such as 鈥渆xtreme catch-up situations鈥 after a long holiday, says Simon Parnall of pay-TV systems maker NDS of Staines, UK. But he questions its overall usefulness. 鈥淚 frankly wonder about the character of a viewer who would wish to use this when up to date with a series,鈥 he says.

Coming soon鈥 talking to your TV

Flat-screen TVs that change channels and volume in response to limited gesture and voice controls are already available from firms like Samsung. But smart TVs will soon get a whole lot smarter, says John Kempner, video buyer at UK chain store John Lewis.

Starting next April, we will be able to ask the next generation of internet-connected TVs questions like 鈥渁re there any movies starring Brad Pitt?鈥 or 鈥渨hat鈥檚 good on TV today?鈥, Kempner says.

Learning software built into the TV scours your viewing history to make recommendations, Kempner says. 鈥淚t will recognise your voice and will display programme choices you should appreciate.鈥