LockPic: Privacy Preserving Photo Sharing in Social Networks

Abstract

There are many privacy concerns related to the use of social networks, in particular the posting of pictures and controlling who has access to them. In this paper we introduce a solution for the distribution of personal or sensitive pictures. Our aim is to provide a method for secure and privacy friendly picture sharing through social networks, that allows users to encrypt sensitive regions in pictures (particularly, faces) in a reversible, non-intrusive way, leaving the rest of the picture unaltered. This way, any image can be freely published and distributed on any social network, and viewed by as many users as the platform allows, while the protected parts are only accessible with the corresponding key. Once the key for a particular region has been acquired, the receiver of the picture can decrypt this region without downloading any additional information. The core of our proposal is a C library, which efficiently integrates an encryption/decryption algorithm with the encoding/decoding process. We have also released an Android application, LockPic, and a companion key server that showcase all the functionality mentioned in this work.

Appendices

A Managing Encryption/decryption Keys

Apart from providing a proper security level and an efficient implementation, one relevant challenge is to properly manage all the encryption keys used in the system. We propose a centralised approach where all keys are stored in the trusted Key Server.

It is essential that the server is able to uniquely identify images in order to be able to generate unique keys for each picture and region in it. As we have mentioned, the Key Server randomly generates a unique identifier for each protected picture that is sent back to the LockPic application at encryption time. This unique ID is included in the metadata of the encrypted picture. Another approach could be to use the hash of the picture as ID. The problem of using the hash as the ID is that the hash has to be performed in the mobile application, which might be an expensive operation depending on the size of the picture, and could present security problems in the case that hash collisions are found. More importantly, the key server would be able to analyse some usage patterns as it would be able to recognize if two different users encrypt the same picture.

As mentioned, the key generation process is performed on the server side. Our initial approach was to generate a separate key for each protected region in every image. This, however, posed some problems because, due to the speed at which random numbers may be needed, the Random Number Generator (RNG) might act as a bottleneck. It would also be difficult to estimate the size of the key store as it would grow in proportion to the number of regions protected. Since having different keys for different regions is mandatory in order to allow for fine grain access control to regions, we have taken the following approach.

For each user, U, a master secret, $MS_U$ is randomly generated at the first access. For every region to encrypt, this secret is concatenated with the picture identifier, ID, and the coordinates of the region, $r=\{x_0,y_0,x_1,y_1\}$; a secure hash function is subsequently applied on this string of bits, and its output is used as the encryption key for the region, i.e.

$$key_{U,ID,r}=hash(MS_{U}\parallel ID \parallel x_0 \parallel y_0\parallel x_1\parallel y_1)$$

The main advantage of this design is that it only uses the RNG once per user and that the number of keys managed by the Key Servers is linear on the number of users, thus independent from the number of pictures or encrypted boxes.

B The LockPic App

The LockPic App uses a very simple user interface with three different choices: Encrypt, Decrypt and My Pictures. The first choice triggers the encryption mechanisms, users are prompted to choose a picture from the gallery and are required to select which regions need to be protected. The selection of protected (Fig. 4a) areas can be performed manually, by placing a box over the desired regions and scaling it by dragging the lower-right corner. Another option is to rely on Android face detection APIs in order to get boxes over the detected faces. In any case, boxes can be easily rearranged and scaled with one finger movement.

Fig. 4.

LockPic user interface

Once the regions have been selected, the user is prompted to select which contacts are authorized to decrypt each of the regions. This step can be skipped and new permissions can be set up later on. Then, the encrypted image (Fig. 4b) that will be stored in the LockPic folder is shown.

Decryption is performed by checking the picture ID included in the metadata and requesting from the key server the corresponding decryption keys. The decrypted image is shown to the user but never stored in the file system. LockPict also provides users with the opportunity to review their access control policies (Fig. 4c). It retrieves from the key server all picture IDs created by the user together with their associated encrypted regions and the list of authorized users and gives the user the choice to modify (add or remove) the users allowed to view each of the regions.